KR970004927B1 - Adaptive motion decision system for digital image stabilizer based on edge pattern matching - Google Patents

Abstract

An adaptive moving vector decision method and an apparatus where decides the figure of subject and the environment of image in the image stabilization system, uses the correlation about the block matching, and stabilizes the images uniformly regardless of the moving frequency in the prescribed area.

Description

Adaptive Motion Vector Determination Method and Apparatus of Digital Image Stabilization System

1 is a block diagram of a first embodiment of an image stabilization system according to the present invention;

FIG. 2 converts the image signal received in FIG. 1 into binary edges, calculates the correlation by matching the pattern of the binary edge signal of the current field and the binary edge signal of the previous field in units of local motion detection. The local motion vector generator generates local motion vectors and statistical variables in each local motion detection area using the calculated correlation.

FIG. 3A is a diagram for explaining a process of generating correlation data by pattern matching a binary edge signal in units of a local motion detection area in FIG. 2, and FIG. 3b shows each local motion detection area in FIG. A diagram illustrating the shape of a pixel to be described.

FIG. 4 is a diagram showing the configuration of the previous field motion detection area storage unit and the edge pattern matching unit in FIG.

FIG. 5 illustrates the correlation between each local motion detection region in FIG. Configuration diagram of the judging unit.

6 is a block diagram of a field motion vector generator for calculating the weight of each local motion detection region using statistical variables in FIG. 1 and generating a field motion vector using the corresponding weights and local motion vectors of each local motion detection region. .

FIG. 7 is a block diagram of a weight generation unit that calculates weights of isolation and stability of local motion vectors of the local motion detection region in FIG. .

FIG. 8 shows candidate motion vectors and correlation diagrams for explaining the process of generating statistical variables and local motion vectors using the correlation coefficients in the local motion decision unit of FIG. Figure of the liver.

FIG. 9a is a characteristic diagram in which the weighting unit of the weight generating unit of FIG. 7 assigns the weight of the isolation degree to the similarity relationship between the local motion vectors. FIG. Characteristic chart that judges the relationship and weights stability.

FIG. 10 is a field for generating a field motion vector by multiplying local motion vectors and weight signals detected from each local motion detection region in FIG. 6 by adding weighted values as a result of adding weights. Configuration diagram of the motion vector determiner.

11 is a cumulative motion vector generator for attenuating the cumulative motion vector by selecting an appropriate attenuation coefficient according to the identification of the panning presence and the cumulative motion vector in FIG. Diagram.

FIG. 12 is a view for explaining a process of selecting an appropriate attenuation coefficient according to the state of the previous cumulative motion vector and the presence or absence of panning of the cumulative motion vector generating part of FIG.

FIG. 13 is a block diagram of a second embodiment of an image stabilization system according to the present invention modified by inserting a compensation cumulative motion vector generator between a cumulative motion vector generator and an address control and zoom processor in the first embodiment.

FIG. 14 is a configuration diagram of a compensation cumulative motion vector generating unit for removing fine error motion by determining a correlation between a cumulative motion vector and an average maximum correlation coefficient in FIG. 13.

FIG. 15A is a diagram illustrating a correlation characteristic between an accumulated motion vector and an error vector compensation value processed by the first error vector compensation determiner in FIG. 14, and FIG. 15B is an average maximum correlation difference and error in the second error vector compensation determiner. A diagram showing the correlation characteristics between vector compensation values.

FIG. 16 is a configuration diagram of a third embodiment modified by inserting a compensation field motion vector generator between a field motion vector generator and a cumulative motion vector generator in the first embodiment.

FIG. 17 is a block diagram of a compensation field motion vector generator for removing fine error motions by determining a correlation between a previous cumulative motion vector and an average maximum correlation difference in FIG. 16.

FIG. 18 is a configuration diagram of a cumulative motion vector generating unit for attenuating a cumulative motion vector and accumulating a cumulative motion vector and a compensation field motion vector by selecting an appropriate attenuation coefficient according to a panning presence identification and a cumulative motion vector in FIG. 16.

* Explanation of symbols for main parts of the drawings

11: local motion vector generator 12: field motion vector generator

13: cumulative motion vector generator 14: address control and zoom processor

15: field memory 21: edge detector

22: previous field motion detection area storage unit 23: edge pattern matching unit

24: local moving judgment 61: weight generator

62: field motion vector determiner 16: compensation cumulative motion vector generator

17: compensation field motion vector generator

The present invention relates to an apparatus and method for stabilizing an image, and more particularly, to an apparatus and method for adaptively determining a motion vector for image stabilization in an image stabilization system.

In general, video cameras are being developed in recent years, with the preference for multifunction equipped with small size, light weight, high magnification of digital zoom and special effects. In this case, when the camera is taken, the shaking of the image is accompanied by the shaking, and when the remote subject is enlarged by the optical or digital zooming, the shaking is amplified, and the stabilized image is provided even in an unstable environment such as the inside of the vehicle. For this reason, the correction function by the Digital Image Stabilization System (hereinafter referred to as DIS system) has emerged as an essential element.

The concept of the image stabilization system as described above is to determine a motion vector between two consecutive video frames in time, and then stabilize the shaking image by moving the position of the following image frame in the reverse direction of the previously detected motion vector. A conventional image stabilization method for such a function has been disclosed by Oshima et al. (M. Oshima, et. Al., VHS Camcorder with Electronic Image Stabilizer) to realize image stabilization by an angular velocity sensor by control of a mechanical imaging optical system. , IEEE Tran. Consumer Elec., Vol. 35, no.4, pp. 749-758, November 1989) .Oshima's image stabilization method uses a gyro sensor to detect an angular velocity of an unwanted camera. Compensate for variations in the image caused by the rotation of the unit. However, the image stabilization method as described above has a problem that is unsuitable for the trend of products requiring small size and light weight due to the increase of the correction mechanism.

The DIS system detects image stabilization only by processing digital signals of an image, and performs image stabilization by LSI technology that identifies image characteristics and adaptively controls the system. In order to implement the DIS system as described above, it is important to have a motion estimation that determines the motion of an image by a minimum of hardware and an algorithm that can adaptively determine a motion vector even under varying conditions. The motion determination unit is generally based on a block matching algorithm (BMA) proposed by Paik et al. edge matching) technology (JKPaik, YCPark, SWPark, An Edge Detectio Approach to Digita1 Image Stabilization Based on Tri-State Linear Neurons IEEE Trans.Consumer Elec., Vol.37.no.3, pp.521-530, August 1991.) and BERP (Band Extract representative Pont) matching technology proposed by Umori et al. (K. Umori, et. Al., Automatic Image Stabilizing System by Full-Digital Signal Processing, IEEE Trans, Consumer Electronics, vol. 36 , no. 3, pp. 510-519, August 1990.) and the three-step search method proposed by Komarek et al. (T.Komarek, et. al., VLSI Architecture for Hierachichical Block Matching Algorithms, IEEE). Trans.consumer Elec., Pp. 45-48, August 1990.

In general, the DIS system compensates for the movement of the entire screen with the motion vectors detected in each field. Due to the characteristics of the actual image, it is expected that the movement by the main subject will exist in the center region. Has Each local motion vector is obtained from these M motion detection regions by a motion vector determination unit, and their various types of correlation data are input to the motion determination unit. Since the local motion vectors determined as described above can be detected differently according to the state of the subject, the motion determination unit determines the field motion vector representing the optimal motion of the video camera adaptively to the image state. Umori proposes a method of processing by a microcomputer using a correlation and a local motion vector to implement the motion determination unit as described above.

However, the method proposed by Sangji Umori has a problem of implementing it in pure hardware because it uses a microcomputer. However, when the edge matching technique is used to determine the motion, the motion decision can be implemented in pure hardware, and thus the processing speed can be very high.

The conventional method of stabilizing an image in the image stabilization system as described above divides the entire screen into a plurality of small screens, and determines the overall motion vector by combining the corresponding local motion vectors obtained from each of them in a suitable manner. However, in the above method, the motion vector is erroneously detected due to the irregular environment of the image and the noise, resulting in instability of the whole system due to the instability of the entire system. As described above, unwanted image conditions include images with low contrast, images with repetitive shapes, images with moving objects, and two or more local motion vectors emitted when intentional panning of the camera occurs. There are images, and images in which a motion vector changes minutely due to noise. Therefore, the image stabilization system should be able to stabilize the image by adaptively determining the motion vector even when the unwanted image is generated.

Accordingly, an object of the present invention is to provide a method and apparatus for adaptively determining a motion vector by determining an environment of an image and a shape of a subject in an image stabilization system.

Another object of the present invention is to provide an adaptive motion vector determination method and apparatus using correlation by block matching in an image stabilization system.

It is still another object of the present invention to provide an apparatus and a method capable of stabilizing an image continuously and uniformly regardless of a moving frequency within a range in which a motion does not exceed a preset range in an image stabilization system.

It is still another object of the present invention to provide an apparatus and method for stabilizing an image in response to an intentional panning of a camera in an image stabilization system.

Still another object of the present invention is to provide an apparatus and method for stabilizing an image having low contrast in an image stabilization system.

Another object of the present invention is to provide an apparatus and method for stabilizing an image having a repetitive form in an image stabilization system.

Still another object of the present invention is to provide an apparatus and method for stabilizing an image having a moving object in an image stabilization system.

It is still another object of the present invention to provide an apparatus and method for stabilizing an image in which a motion vector changes in time by noise in an image stabilization system.

In order to achieve the above object, the present invention receives digital image data generated from a camera, converts received image data into binary edge data, and detects local motion by matching patterns of edge data of the current field and the previous field. Means for sequentially generating the correlation data of the area units, analyzing the generated correlation data, and generating statistical variables and local motion vectors of the corresponding local motion detection areas sequentially, and the locally generated motion vector And statistical variables, calculates the weight by calculating the isolation and stability of the local position vector, and calculates the field motion vector by increasing the calculated weights and the corresponding local motion vectors at field periods. And attenuating means and said field motion vector and correlation. Receives and identifies the presence or absence of intentional panning from the field motion vector, and selects the attenuation value of the attenuation means in accordance with the presence or absence of the panning identification signal PID, and the previous cumulative motion vector for outputting the attenuation means Means for accumulating the received field motion vector as the cumulative motion vector and applying it to the attenuation means, a memory for receiving and storing the projection data in units of fields, and receiving the cumulative motion vector, The controller is configured to control a stable image screen by adaptively correcting a motion vector included in the image data.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

It should be noted that the same elements in the figures are represented by the same numerals wherever possible. In the following description, the contents described as specific numerical data as examples are provided to help general understanding of the present invention. The present invention has been described assuming a process of dividing an area of one field into four local motion detection areas. Therefore, it will be apparent to those skilled in the art that these data can be modified to practice the present invention.

1 is a configuration diagram of a first embodiment of an image stabilization system according to the present invention, wherein digital image data generated from a camera is applied to a local motion vector generation unit 11 and a field memory 15. The local big vector generator 11 receives digital image data, detects a binary edge signal of the current field from the received image data, and converts the detected binary edge signal of the current field into a binary edge of the previous field. Block matching of the signal with the unit of signal and local motion detection area to calculate the correlation by comparing two consecutive fields, and using this correlation data, the local motion vector (LMV) of the corresponding local motion detection area. And statistical variables. The field motion vector generator 12 is connected to the output terminal of the local motion big vector generator 11 to receive local motion vector LVMs and statistical variables. The field motion vector generation unit 12 generates a field motion vector (FMV) representing the motion of one complete field from the received local motion vector LMVs and statistical variables. The cumulative motion vector generator 13 is connected to the output terminal of the field motion vector generator 12 and accumulates the received field motion vector FMV to stabilize the shaking between successive fields to an initial state. Generates (Accumulated Motion Vector: AMV). The field memory 15 receives and stores the image data, and outputs the image data stored in the corresponding area by the read address. The address control and zoom processing unit 14 generates a read address from the cumulative motion vector AMV received from the cumulative motion vector generator 13 and applies the read address to the field memory 15 and is read from the field memory 15. A digital zooming process is performed to receive image data and to compensate for margin data of the outer portion generated by motion compensation. That is, the address control and zoom processing unit 14 digitally zooms the image data received from the field memory 15 by the read address to compensate for the margin data of the outer portion by the motion compensation through the above process. The digital zoom process interpolates an image signal, enlarges a predetermined portion of the image, and finally outputs a stabilized image.

In the image stabilization system as shown in FIG. 1, in order to determine a motion vector between two consecutive images, which are the reference image and the comparison image, the local motion vector generating unit 11 first moves the appropriate position of the received image data. Set in the Motion Estimation Area (MEA). Thereafter, the local motion vector generation unit 11 compares the image data of the local motion detection region in which the reference image exists with the image data of the local motion detection region in which the comparison image exists, and compares each of the local motion detection regions. The images are shifted by the corresponding local motion vector candidate. When comparing the image data of the reference local motion detection region and the comparison local motion detection region, a correlation value is assigned to corresponding motion vector candidates, respectively. A motion vector candidate having the maximum correlation among a plurality of motion vector candidates is selected and output as a local motion vector. In addition, the local motion vector generation unit 11 determines M local motion vectors different from each other by M local motion detection regions per unit field in order of time in order to more reliably determine the motion vector. In determining the local motion vector as described above, the present invention uses binary edge image data instead of using actual image data or filtered multi-bit image data in order to reduce the number of hardware of the calculation and storage means. . Therefore, by using the binary edge pattern matching technique as described above in case of local motion vector LMV generation, the hardware of the image stabilization system can be greatly reduced.If the movement of the image data does not exceed the set range, it is continuous regardless of the movement frequency There is an advantage that can uniformly stabilize the image data.

By properly combining the local motion vector LMVs generated as described above, the field motion vector FMV can be obtained. To this end, the field motion vector generator 12 receives M local motion vectorers from the local motion vector generator 11 to obtain a field motion vector FMV which is a desired motion vector between the reference field and the comparison field. In order to generate the field motion vector as described above, the field motion vector generator 12 adaptively determines weights for the isolation and the stability of each local motion vector, and determines as described above. The weights are combined with the local motion vector. In addition, the cumulative motion vector generator 13 receiving the output of the field motion vector generator 12 accumulates successively received field motion vectors FMVs in order to more stabilize image data shaking between two consecutive fields. Generates a cumulative motion vector AMV. The address control and zoom processor 14 receives the cumulative motion vector AMV output from the cumulative motion vector generator 13 and controls the output of the image data stored in the field memo 15 as described above. To compensate. That is, the addresser and the zoom processor 14 receive the cumulative motion vector of the cumulative motion vector generator 13 to calculate the read address of the field memory 15, and the field memory 15 corresponds to the corresponding area. Outputs video data stored in. At this time, the address control and zoom processor 14 compensates the margin data of the outer portion generated in the movement of the image data, which is finally stabilized by zooming a certain portion of the image data received by the interpolation method. Will print

As described above, the image stabilization system of the present invention has a local motion vector generator 11, a field motion vector generator 12, a cumulative motion vector generator 13, address control and zoom as shown in FIG. It has five structural units which consist of the processing part 14 and the field memory 15. FIG. The component unit adaptively compensates image data according to instability conditions of variously generated image data. The unstable condition is a noise component in a motion vector when the screen is moved due to deliberate panning, when a low contrast image is received, when the image has a repetitive shape, or when a moving object exists. This is the case with the existing video. In the above case, the deliberate panning is an unstable condition for reducing the cumulative motion vector AMV, and the rest are unstable conditions for lowering the local motion vector LMV.

The operation of the first embodiment of the image stabilization system of the present invention having the configuration as shown in FIG. 1 will be described.

2 is a block diagram of the local motion vector generator 11 of FIG. 1, wherein the edge detector 21 receives the image data, and receives binary edge data from the image data of the current field, which is the actual image data received. edge data) is detected and output. The construction and operation of the edge detection 21 is described in detail in patent application 91-4871 filed by the applicant of the present application. The edge detection unit 21 rests the binary edge data to the edge pattern matching unit 23 and the previous field motion detection area storage unit 22, respectively. The previous field frame detection area storage unit 22 includes M local motion detection areas, and scans the binary edge data of the current field received from the edge detection unit 21 in units of local motion detection areas. And store it in the corresponding local motion detection area. Therefore, the previous field motion detection area storage unit 22 delays the binary edge data stored in each local motion detection area for one field period, and the edge at the time of detecting the local motion vector LMV from the corresponding local motion detection area. Output to the pattern matching unit 23. The edge pattern matching unit 23 receives the binary edge data of the current field output from the edge detection unit 21, and transfers the previous field of the corresponding local motion detection area from the previous field local motion detection area storage unit 22, respectively. Receive binary edge data. The edge pattern matching unit 23 pattern-matches the binary edge data of the previous field in addition to the current binary edge data in the unit of the local motion detection area and outputs M phase data CORs. The construction and operation of the previous field local motion detection area storage section 22 and the edge pattern matching section 23 are described in detail in patent application No. 91-10201 filed by the present applicant. The local motion determining unit 24 receives the output of the edge pattern matching unit 23 and generates local motion vector LVMs of respective corresponding local motion detection regions from the received correlation CORs.

FIG. 3A is a diagram illustrating a process of generating a local motion vector of a region corresponding to a local motion detection region after converting image data of one field into a binary edge signal according to the present invention. FIG. The example which comprises a detection area is shown. In addition, each local motion detection area has a search motion estimation area (SMEA) centered on a motion vector estimation area (MEA), and two comparison motions within the search motion decision area SMEA. Compared Motion Vector Estimation Area (CMEA). 3B is a diagram showing the number of image data of the reference motion vector detection area MEA and the comparison motion vector detection area CMEA. In the present invention, the number of constituent pixels of the motion vector detection area is assumed to be 32 * 9. The pixel in the direction is selected by selecting one of two pixels in the odd or even pixel from 63 pixels, and the pixel in the column direction selects one pixel every four lines. Therefore, the pixel of the motion vector detection area requires 63 pixels in the row direction and 31 lines in the column direction.

FIG. 3C shows various timing signals used in the present invention. VD is a vertical synchronization signal, and MDCK is a local motion vector generated at a point when the search of the local motion detection region is terminated as shown in FIG. 3A during one field period. Sampling clock, SECLR is a clear signal for clearing the information in the previous local motion detection area after the MDCK generation and to detect the information of the next local motion detection area, FCK is a field motion for generating the field motion vector FMV A sampling clock of a vector and ACK is a sampling clock of a cumulative motion vector generating the cumulative motion vector AMV. Also, between the SECLR and the FCK, the field motion vector FMV is generated by sequentially processing the information of the local motion detection areas that have been generated and stored. SCK is a sampling clock of image data, and LMVS0 and LMVS1 are field motion vector FMV. A control signal for sequentially selecting each local motion vector LMV and a weight signal corresponding to each occurrence, and WTCK is a clock for multiplying and adding a local signal vector and a weight signal corresponding to each local motion vector.

4 shows the structure of the previous field motion detection area storage 22 and the edge pattern matching section 23 in FIG. Here, the previous field motion detection area storage unit 22 should be able to store data of the reference motion vector area MEA of each local motion detection area in the image of one field. Accordingly, the reference data extractor 41 extracts binary edge signals corresponding to the position of the reference motion vector detection region MEA of each local motion detection region from the binary edge signal of one field output from the edge detection unit 21. . The binary edge signal outputting the reference data extractor 41 becomes reference area data consisting of 32 * 9 = 288. The reference data storage unit 42, which receives the output of the reference data extraction unit 41, stores the reference data generated from the reference motion vector detection areas MEAs. Accordingly, the reference data storage unit 42 is configured to correspond to the number of local motion detection areas, and shifts and stores the received reference area data whenever the local motion detection areas are moved. The reference data storage unit 42 outputs the reference area data to the edge pattern matching unit 23 at a point of time delayed by one field period.

In the edge pattern matching unit 23, the comparison data extracting unit 43 receives the output of the edge pattern matching unit 23, and outputs the received binary edge signal of area 3b and 32 * 9 (288). Occurs. The edge pattern comparison unit 44 receives the output of the comparison data extracting unit 43 as comparison area data and receives the output of the reference data storage unit 42 as reference area data. The edge pattern comparison unit 44 compares the edge patterns of the two pieces of data in pixel units and outputs the pixels. The correlation generator 29 receives the output of the edge pattern comparison unit 44 and calculates the number of pixels having the same logic from the comparison signals of the compared pixels to generate the correlation data COR. Since the block data is composed of 288 pixels, the correlation data COR is generated with 9 bits of data.

FIG. 5 is a configuration diagram of the local motion decision unit 24 of FIG. 2 and shows a configuration of the motion determination unit 24 generating local motion vector LVMs. Looking at the configuration of the local motion determination unit 24, the received correlation data COR is a correlation diagram sequentially generated according to the position of the search movement detection areas from the edge pattern matching unit 23. The local motion determining unit 24 receives the correlation to generate a local motion vector, a maximum correlation degree Cmax, a second maximum correlation degree C _2nd , a maximum correlation difference Cdif, and an average correlation degree Cavg. The means for generating the maximum correlation degree Cmax is composed of registers 501, 502, 503, comparator 51 and multiplexer 52. The register 501 receives and stores the correlation diagram COR of the corresponding local motion detection region generated from the edge pattern matching unit 23. The comparator 51 receives the correlation data COR and the previous maximum correlation degree Cmax, and compares the magnitudes of the two received signals to generate a comparison result signal. The comparison result signal is a 1-bit signal indicating whether the current correlation data COR is greater than or less than the previous maximum correlation degree Cmax. The register 504 receives the output of the comparator 51 and stores the received comparison result signal. The multiplexer 52 receives a current correlation diagram COR which receives the previous maximum correlation degree Cmax as the first terminal and outputs the register 501 as the second terminal, and outputs from the register 504 as a selection terminal. Receive a comparison result signal. The multiplexer 52 selectively outputs the previous maximum correlation value Cmax received at the first terminal according to the logic of the comparison result signal or changes the current correlation data COR received at the second terminal to the maximum correlation degree Cmax. Output The register 502 stores the maximum correlation degree Cmax output from the multiplexer 52. The means for generating the maximum correlation degree Cmax compares the current correlation degree data COR received from the edge pattern matching unit 23 with the maximum correlation degree Cmax in the previous state, and the phase correlation data having a large correlation value as a result of the comparison. Is stored at the same time as the maximum correlation degree Cmax. The means for generating the second maximum correlation C ₂ nd consists of an AND gate 53 and a register 503. The AND gate 53 receives the output of the register 504 and the system clock SCK, and controls the output of the system clock SCK according to the logic of the comparison result signal. The register 503 receives the previous maximum correlation diagram Cmax, and outputs the previous maximum correlation diagram Cmax received when the system clock SCK is received as the second maximum correlation diagram C _2nd . It means for generating a second maximum correlation C ₂ nd is also stored in the same oi outputting a C ₂ nd the maximum correlation Cmax second maximum correlation in the previous state when the change is also the maximum correlation Cmax. The means for generating the maximum correlation difference Cdif is a subtractor 54, which is generated as the maximum correlation difference Cdif by subtracting the second maximum correlation degree C _2nd from the maximum correlation degree Cmax. The means for generating the average correlation Cavg is composed of an adder 56, a register 506 and a divider 57. The adder 56 adds the current correlation diagram COR received from the edge pattern matching unit 23 and the accumulated correlation diagram data. The register 506 receives and stores the output of the adder 56 and simultaneously applies accumulated correlation data to the adder 56. The divider 57 receives the output of the register 506 and divides by the number of candidate motion vectors to generate an average correlation Cavg. The means for generating the average correlation diagram Cavg successively adds the received correlation diagram COR to the correlation diagram data corresponding to the candidate motion vector, and divides the cumulative accumulated correlation by the number of candidate motion vectors to generate the average correlation diagram Cavg. do. The means for detecting a local motion vector is composed of a register 505 and a local motion vector detection unit 55. The register 505 stores the output of the register 504. The local motion vector detector 55 calculates the position of the current correlation data COR by counting the received address clock, and when the maximum correlation degree Cmax is updated according to the logic of the comparison result signal output from the register 505. It is used as the location information of the current correlation diagram COR at this point of time and is detected by the local motion vector LVM at the final correlation position.

FIG. 6 is a block diagram of the field motion vector generator 12 generating the field motion vector FMV by receiving the statistical variables and the local motion vector LVMs output from the local motion decision unit 24 of FIGS. 2 and 5. to be. The local motion decision unit 24 receives the correlation data COR of each local motion areas generated from the edge pattern matching unit 23 as described above, and generates local motion vector LVMs and statistical variables. The field delay unit 63 receives the field motion vector FWV and generates a previous field motion vector (PFMV) delayed by one field period. The average local motion vector generator 64 adds and divides local motion vector LVMs received from the local motion decision units 24 to generate an average local motion vector (ALMV). The average maximum correlation difference generation unit 34 adds and divides the maximum correlation difference Cdif received from the local motion determination unit 24 to generate a correlation maximum difference Cad (Correlation Average Difference). The weight generator 61 receives a previous field motion vector PFMV, a local motion vector LMV, an average local motion vector ALMV, an average correlation Cavg, a maximum correlation capital processing, and a second maximum correlation C ₂ nd, and determine the local movement. A weight signal corresponding to the local motion vector LMV sequentially generated in the unit 24 is continuously generated. The weight generator 61 generates weights for stability and isolation using the local motion vector LMV, the previous field local motion vector PFMV, and the average local motion vector ALMV received from the local motion decision unit 24. After determining whether the video signal is unstable using the statistical variables Cavg, Cmax and C ₂ nd, the calculated weight signal is reset or not. The field motion vector determiner 62 multiplies the localized motion vector LVMs corresponding to the weighted signals received from the weight generator 61 to give weights to the local motion vector LVMs, and is weighted. The local motion vectors are combined to generate the field motion vector FMV.

FIG. 7 is a configuration diagram of the weight generator 61 of the fifth diagram for adaptively processing a local motion vector generated in an irregular image. The means for selectively outputting the local motion vector may include a first storage unit 701 and a first motion vector. And a first selector 702. The first storage unit 701 receives the local motion vector LMV and stores a local motion vector LMV received by the MDCK generated at the end of the search of the local motion detection region. The first storage unit 701 is illustrated assuming an example of generating four local motion vectors LMV. Therefore, the local motion vector LMV is sequentially stored in the register 141-register 144 by the MDCK in the order of occurrence. The first selector 702 receives a local motion vector outputting the first storage unit 701, terminates the search of one field, and sends it to the first storage unit 701 according to the logic of LMVSO and LMVS1. The stored corresponding local motion vector LMV is selected and output as the local motion vector LMV. The means for selectively outputting the average correlation Cavg comprises a second storage unit 703 and a second selector 704. The second storage unit 703 receives the average correlation diagram Cavg, and stores the average correlation diagram Cad received in the MDCK generated at the end of the search of the local motion detection region. The four average correlations of the second storage unit 703 are also shown as an example of generating Cavg. The second selector 704 receives the average correlation degree Cavg received from the second storage unit 703, terminates the search of one field, and generates the second storage unit 703 according to the logic of LMVSO-LMVS1. Select the corresponding mean correlation Cavg stored in to generate the mean correlation Cavg. The means for selectively outputting the maximum correlation degree Cmax is composed of a third storage unit 705 and a third selector 706. The third storage unit 705 receives the upper maximum correlation degree Cmax, and stores the maximum correlation degree Cmax received in the MDCK generated at the end of the search of the local motion detection region. Four maximum correlations of the third storage unit 705 are shown assuming an example of generating Cmax. The third selector 706 receives the maximum correlation degree Cmax received from the third storage unit 705, terminates the search of one field, and generates the third storage unit 705 according to the logic of LMVS0 and LMVS1. Generate the maximum correlation Cmax by selecting the corresponding maximum correlation Cmax stored in. The means for selectively outputting the second maximum correlation C ₂ nd includes a fourth storage unit 707 and a fourth selector 708. The fourth storage unit 707, and the second maximum correlation receiving a C ₂ nd, and the second maximum correlation received by the MDCK generated at the end of browsing of the local motion-detection area also stores the C ₂ nd. Four second maximum correlation diagrams of the fourth storage unit 707 are illustrated assuming an example of generating C _2nd . The fourth selector 708 receives the second maximum correlation diagram C _2nd received from the fourth storage unit 707, and ends the search of one field and stores the fourth storage according to the logic of LMVS 0-LMVS1. The second maximum correlation diagram C _2nd stored in the unit 707 is selected to generate the second maximum correlation diagram C _2nd .

The means for determining the isolation weight is composed of a subtractor 711 and an isolation weight coding unit 713. The subtractor 711 detects an isolation degree by calculating a difference between the average local motion vector ALML and the local motion vector LMV. The isolation weighting coordinate unit 713 receives the isolation output from the subtractor 711, and assigns a weight corresponding to the value of the isolation to generate an isolation weight of the corresponding local motion vector LMV. The means for determining the stability weight value comprises a subtractor 712 and a stability weight value coding portion 714. The subtractor 712 calculates a difference between the local motion vector LMV and the previous field motion vector PFMV to detect stability. The stability weighting coding unit 714 receives the stability from the subtractor 712 and assigns a weight corresponding to the received stability value to generate a stability weight of the corresponding local motion vector LVM. In addition, when the local motion vector is generated from the unstable image data, the field motion vector FMV should be adaptively generated according to this condition. When the local motion vector has a low contrast, the image has a repetitive image shape. And moving objects. First, the function of detecting an image with low contrast is performed by the comparator 723.

The comparator 723 receives the average correlation Cavg and the first threshold REF1, and compares the average correlation Cavg with the first threshold REF1, which is a reference for low contrast, so that the average correlation Cavg is the first threshold REF1. If smaller, it is determined by unstable image data. The second means for detecting the image of the moving object consists of a subtractor 721 and a comparator 724. The subtractor 721 receives the average correlation Cavg and the maximum correlation Cmax and subtracts the average correlation Cavg from the maximum correlation Cmax to generate a difference signal. The comparator 724 receives the output of the subtractor 721 and the second threshold value REF2, and determines that the image signal is unstable when the difference signal is smaller than the second threshold value REF2, which is a reference for determining the moving object. The third means for detecting an image having a repetitive form consists of a subtractor 722 and a comparator 725. The subtractor 722 receives the maximum correlation degree Cmax and the second maximum correlation degree C _2nd, and subtracts the second maximum correlation degree Csnd from the maximum correlation degree Cmax to generate a difference signal. The comparator 725 receives the output of the subtractor 722 and the third threshold value REF3, and determines that the image signal is unstable when the difference signal is smaller than the third threshold value REF3, which is a reference for determining a repetitive image. The means for generating the adaptive local motion vector consists of an adder 715, a divider 716, a register 717 and an endgate 726. The adder 715 receives and adds the isolated and stable weights. The divider 716 receives the output of the adder 715 and averages the two added weights to generate a weight W of the corresponding local motion vector LMV. The register 717 stores the averaged weight W.

The AND gate 726 receives the outputs of the comparators 73, 74, and 75, and applies the output control signal of the register 717. The register 717 stores the averaged weights and controls the output of the averaged weights W stored when the signal is received as an unstable image signal from the AND gate 726. Therefore, the means for generating the adaptive local motion vector is averaging of the corresponding local motion vector LMV when the received image signal is a video signal having a low contrast, a repetitive image, or a moving object. The weight W is controlled to be reset, and the weight W is output.

8 is a characteristic diagram showing a relationship between correlation data COR and video data. In the case of (6a) and the image data having normal contrast, the range of the correlation has a large range, and in the case of the image having a low contrast as shown in (6b), the local motion vector LMV detected by decreasing the correlated average value. Falls accuracy. In addition, in the repetitive image as shown in (6c), the difference in correlation becomes smooth.

9a shows the isolation degree by calculating the difference between the local motion vector LMV and the average local motion vector ALMV, and then isolates the corresponding local motion vector LMV according to the value of the isolation degree obtained from the isolation weighting coding unit 713. A characteristic diagram giving the weighting value is shown. Here, the isolation degree is a correlation deviation amount of each local motion vector LMV. 9b shows the stability of the local motion vector LMV according to the value of the stability obtained from the stability weight coding unit 714 after calculating the difference between the local motion vector LMV and the previous field motion vector PFMV. The characteristic diagram to be provided is shown. Here, stability means the amount of deviation from the entire local motion vector LMV.

FIG. 10 is a configuration diagram of the field motion vector determining unit 62 in FIG. 5, wherein the weighting means is composed of a multiplier 101 and an adder 102. As shown in FIG. The multiplier 101 sequentially multiplies the local motion vectors sequentially output from the first selector 702 of the weight generator 61 with the weight signals corresponding to the local motion vector LMV, and weights each local motion vector. Grant. The adder 102 receives the output of the multiplier 101 and adds it sequentially to generate a motion vector corresponding to one field period, and applies it as a dividend. The weight counting means is performed by the adder 103. The adder 103 adds the received weight signals and applies them as a divisor. The adder 103 cannot perform the addition function when the weight signal is 0, and as a result, only the reliable local motion vector LMV is multiplied with the weight and outputted in the corresponding field period. The field motion vector calculation means is performed by the divider 104. The divider 104 receives the output of the accumulator 102 as the dividend, receives the output of the adder 103 as the divisor, and generates a field motion vector FMV by the division process.

11 is a configuration diagram of the cumulative motion vector generating unit 13 in FIG. Is generated as intentional panning, and generates a panning identification signal PID for changing the attenuation coefficient of the cumulative motion vector AMV.

In addition, the comparator 119 receives a cumulative motion vector AMV up to a comparison input and receives a fourth threshold value REF 4 as a reference input so that the magnitude of the cumulative motion vector AMV does not exceed the maximum correction region. The comparator 119 generates a comparison result signal for changing the attenuation coefficient of the cumulative motion vector AMV when the magnitude of the cumulative motion vector AMV is larger than the fourth threshold value REF4. The oragate 120 receives the outputs of the panning identifier 111 and the comparator 119 and generates the received signal as a selection signal for selecting an attenuation coefficient. The attenuation determining means is composed of a multiplexer 118 and attenuators 116 and 117. The first attenuator 116 receives the cumulative motion vector AMV up to the previous time, and attenuates the cumulative motion vector AMV received with the set first attenuation coefficient K1. The second attenuator 117 receives the cumulative motion vector AMV up to the previous, and attenuates the cumulative motion vector AMV received with the second attenuation coefficient K2. The multiplexer 118 receives the output of the first attenuator 116 as a first terminal, receives the output of the second attenuator 117 as a second terminal, and outputs the output of the oragate 120 to a selection terminal. Receive. The multiplexer 118 selectively outputs the output of the second attenuator 117 through which the panning identification signal PID or the comparison result signal is received, and otherwise outputs the output of the first attenuator 116. The cumulative motion vector generating means is composed of registers 112 and 115, an adder 113 and a limiter 114. The register 112 stores the received field motion vector FMV at the time of FCK generation. The adder 113 receives the outputs of the register 112 and the multiplexer 118 and adds the two signals to generate a cumulative motion vector AMV. The limiter 114 receives the output of the adder 113 and limits when the cumulative motion vector AMV is more than a predetermined size. The register 115 receives the output of the limiter 114 and stores it when the ACK is generated, applies it to the inputs of the attenuators 116 and 117 and the comparator 119 and outputs the same to the address control and zoom processor 14.

FIG. 12a shows the timing of the panning identification signal PID in the panning identification unit 111. FIG. 12b shows the movement amount of the cumulative motion vector AMV according to whether the panning identification signal PID and the comparator 119 are generated. FIG. 12C illustrates a process in which attenuation coefficients are compensated to generate a cumulative motion vector AMV as shown in FIG. 12B when panning is detected from the field motion vector FMV. In the drawing, the cycle T1-T2 illustrates a process of outputting the cumulative motion vector AMV obtained by attenuating the first damping coefficient K1 by generating the cyst identification signal PID, and the cycle T3-T4 represents the magnitude of the cumulative motion vector AMV. Shows a process of outputting a cumulative motion vector AMV obtained by attenuating the first attenuation coefficient K1 in a state where is larger than a fourth threshold value, which is an upper limit of correction.

The operation of the first embodiment of the image stabilization system according to the present invention will be described in detail with reference to FIGS.

The local motion vector generation unit 11 shown in FIG. 2 receives real image data of one field, converts it into binary edge pattern data, and outputs M local motion data of the edge pattern data. Correlation data COR is generated by pattern matching the binary edge data of the previous field divided into regions. The M correlation data CORs are respectively received by the corresponding local motion determination unit 24 to generate a local motion vector LMV in a corresponding local motion detection region, wherein the M local motion vectors LMV are field motion vectors and As it is used to generate the cumulative motion vector AMV, it plays a very important role in the image stabilization system. Therefore, if the local motion vector LMVs are generated correctly, the field motion vector FMV and the cumulative motion vector AMV only need to be averaged and accumulated.

The local motion vector generator 11 detects an edge of the actual image data received by the edge detector 21 and converts it into a binary edge signal as shown in FIG.

The edge signal output from the edge detector 21 is a binary signal of 1 bit. Then, the previous field motion detection area storage unit 22 reference data extractor 41 receives binary edge data outputted to the edge detection unit 21 to detect the reference motion vector in each local motion detection area as shown in FIG. 3a. The reference blocks of the area MEA are extracted. The reference data storage unit 42 stores the reference block data at the end of each local motion detection area, and shifts the reference block data of the previous local motion detection area previously extracted and stored. Accordingly, the reference data storage unit 42 delays the received reference block data by one field period, and then outputs the binary edge data of the current field of the corresponding local motion detection region in the corresponding local motion detection region of the previous field. The extracted reference block data is output. The comparison data extracting unit 43 receives the binary edge data and generates comparison block data as shown in FIG. 3B. Then, the edge pattern comparison unit 44 compares the block pattern of the comparison block data with the reference block data to determine the number of edge data having the same logic in the corresponding local motion detection area of the previous field and the current field. In this case, the motion vector detection region having the greatest correlation with the reference block data in FIG. 3a will be the comparison motion vector detection region CMEA. The correlation generator 29 receiving the output of the edge pattern comparison unit 44 receives the comparison result signal, calculates the number of pixels having the same logic, and generates the result as the correlation data COR. Therefore, it can be seen that the correlation data COR is the number of binary edge data matched in the block data of each local motion detection region in two consecutive temporal fields. Then, the local motion decision unit 24 sequentially generates the local motion vector LMV and statistical variables of the local motion detection regions using the correlation data COR.

First, a process of adaptively detecting a motion vector using the correlation data CORs will be described. The local motion determination unit 24 receives the correlation diagram COR of the corresponding local motion detection area to generate a local motion vector LMV, and is used by the weight generator 61 and the field motion vector determiner 62. Obtain statistical variables. The statistical variables may include a maximum correlation degree Cmax, a second maximum correlation degree C _2nd , a maximum correlation degree Cdif, and an average correlation degree Cavg, etc. The field motion vector generation unit 12 generates the local motion vector. Using the local motion vector LMV and the statistical variables Cmax, C ₂ nd, Cdif, and Cavg, which are output from the local motion decision unit 24 of the unit 11, the local motion vector LMV generated in the local motion detection region is obtained, and among them, Selects only reliable local motion vectors LMVs.

In addition, the weight of the local motion vector LMV, which is determined by checking the local motion vector LMV, the previous field motion vector PFMV, and the average local motion vector ALMV, is determined based on the isolation and stability of the local motion vector LMV. These fields are averaged to determine the field motion vector FMV. That is, the weight generator 61 calculates the isolation and stability from the detected local motion vector LMV, the average local motion vector ALMV and the previous field motion vector PFMV, and averages the calculated weights of the isolation and stability. Weighting is applied to the motion vector LMV. From the statistical variables Cmax, C ₂ nd and Cavg, it is examined whether the image having a low contrast, the image having a repetitive shape, the moving object is included, and the reliability of the local motion vector LMV is determined based on the test result. Select only reliable local motion vectors LMV. The field motion vector determiner 62 generates a reliable field motion vector FMV by calculating an average of the weights obtained by the weight generator 61 and the local motion vector LMVs corresponding thereto. The cumulative motion vector generator 13 accumulates the field motion vector FMV determined by the field motion vector generator 12 in time and calculates a cumulative motion vector AMV for stabilizing the shaking between successive fields to an initial state. . The address control and zoom processor 14 then implements a zoom function using the cumulative motion vector AMV.

Referring to FIG. 5, the operation of the local motion determination unit 24 is shown. Among the correlation data CORs generated from the local motion detection areas as shown in FIG. 3a, the correlation data COR of the corresponding local motion detection areas is respectively shown. Receive and generate local motion vector LMV and statistical variables. The received correlation data COR is received from the edge pattern matching unit 23 and is the number of matched binary edge data sequentially obtained in two fields according to the position of the local motion detection region. The registers 501-506 and the local motion detection unit 55 of the local motion determination unit 240 are reset at each occurrence of the SECLR signal generated when the search of the local motion detection area ends, and then the next local motion detection area is generated. Prepare to detect local motion vector of LMV and statistical variables.

First, the generation process of the maximum correlation degree Cmaxm is described. The correlation data COR output from the edge pattern matching unit 23 is applied to one terminal of the comparator 51. At this time, the other terminal of the comparator 51 receives the maximum correlation degree Cmax in the previous state, and has a reset state at the first time when the search for the local motion detection area is started.

Therefore, the fraud comparator 51 compares the currently received correlation data COR with the previous maximum correlation degree Cmax and outputs it to the register 504, which registers it as a selection signal of the multiplexer 52. . At this time, the multiplexer 520 receives the maximum correlation diagram Cmax outputting the register 502 as the first terminal and the current correlation diagram COR received through the register 501 as the second terminal. Therefore, when the maximum correlation degree Cmax value in the comparator 51 is greater than the current correlation data COR value, the first comparison signal is output as a comparison result signal, and thus the multiplexer 52 causes the first terminal. The maximum correlation degree Cmax received as is selected as it is, so that the previous maximum correlation degree Cmax is maintained at the maximum correlation degree Cmax, but the current correlation data COR value in the comparator 51 is the maximum correlation value Cmax. If the value is larger than the value, the second comparison signal is output as a comparison result signal. As a result, the multiplexer 52 selects the current correlation data COR received at the second terminal as the maximum correlation degree Cmax. And in which case the maximum correlation Cmax also failed to determine the changes, and the register 502 receives and stores the maximum correlation Cmax is also outputted from the multiplexer 52.

A second maximum correlation for the second also look at the generation process of the C ₂ nd, and according to the logic of the comparison result signal and outputting the comparator 51 and the second maximum correlation determining a C ₂ nd. That is, the AND gate 53 which receives the comparison result signal cuts the output of the system clock SCK when the first comparison signal is received, and maintains the value of the register 5030 as it is. maximum correlation of the previous Figure is to maintain the previous second maximum correlation as a C ₂ nd to be smaller than Cmax value, but when receiving the second comparison signal from the register 504, the maximum correlation Cmax is changed The AND gate 53 outputs the system clock SCK. Therefore, the register 503 for receiving the previous maximum phase Cmax and receiving the output of the AND gate 53 as a clock is the system clock SCK. By storing the previous maximum correlation Cmax as the second maximum correlation C ₂ nd value.

Accordingly, it can be seen that the second maximum correlation degree C _2nd value becomes the maximum correlation degree Cmax that was maintained before the change point when the maximum correlation degree Cmax value is changed.

Third, the maximum correlation value Cdif may be subtracted from the maximum correlation value Cmax value output from the register 502 and the second maximum correlation value C _2nd value output from the register 503. This is done in subtractor 54. Fourth, the average correlation Cavg can be divided by a predetermined value after adding the received correlation. In this case, the predetermined value is the number of candidate motion vectors in the local motion detection region. To this end, the adder 56 adds the correlation data COR received from the edge pattern matching unit 23 to the correlation values accumulated and stored in the register 506 before, and the register 504 stores them again. The divider 57 divides the output of the register 504 by a predetermined value. Then, the value outputting the divider 57 becomes the average diffuser Cavg, and the correlation value added to the adder 56 is stored in the register 506 again to prepare for the next state.

Cmax, C _2nd , Cdif, and Cavg generated as described above are obtained statistical variables, and the motion vector detector 55 receives the statistical variables to obtain a local motion vector LMV in the corresponding local motion detection region. Looking at the detection process of the local motion vector LMV, the comparison result signal of the current correlation data COR value outputting the register 504 and the previous maximum correlation degree Cmax value is applied to the register 505 and stored. At this time, the motion vector detector 55 counts the address clock and calculates the position of the corresponding correlation data COR. The address clock is a correlation sampling clock in the candidate motion vector detection area. When the motion vector detector 55 receives a second comparison signal informing the update of the maximum correlation degree Cmax from the register 505, the motion vector detector 55 recognizes this and localizes the position value calculated at the corresponding time point. Output as a motion vector LMV. Therefore, the local motion vector LMV becomes position information of a block having the maximum correlation degree Cmax in the corresponding local motion detection area.

Therefore, the local moving decision unit 24 compares the received correlation data COR value with the previous maximum correlation value Cmax and then maintains the current state as it is when the previous maximum correlation value Cmax is large. The current correlation value is changed to the maximum correlation degree Cmax and the maximum correlation degree Cmax value in the previous state is changed to the second maximum correlation degree C _2nd .

In addition, the motion vector detection unit 55 calculates the position of the current block by using the address clock. When the maximum correlation value Cmax is updated, the motion vector detection unit 55 outputs the current position as a local motion vector LMV. It can be seen that the correlation data having a CORDML position. The maximum correlation value Cavg may be calculated by subtracting the second maximum correlation value C ₂ nd value from the maximum correlation value C max value. The local motion determining unit 24 is reset by the search end signal SEC-CLR generated at the end of the search of each local motion detecting region as shown in FIG. Preparation is made to detect local motion vectors and statistical variables.

The local motion vector LMV generated as described above has a significant effect on reliability. The accuracy of the local motion vector LMV detected as described above is changed according to the correlation data COR (i, j) shifted by (i, j), which is the coordinate of the candidate motion vector of the calculation result of the correlation calculation block. If the correlation diagram data is similar, the reliability of the detected local motion vector LMV should be evaluated by the correlation diagram COR of each local motion detection area, and the field motion vector FMV that represents each field is used in the reliable local motion detection area. The local motion vector LMV obtained should be selected and determined.

Looking at the operation of the field motion vector generator 12 with reference to FIG. 6, the weight generator 61 collects and stores the received local motion vector LMVs and statistical variables in units of the local motion detection area and stores the field motion. At the time of generating the vector FMV, weighted signals are generated by sequentially analyzing the local motion vector LMV, and statistical variables corresponding to the local motion vector LMV are analyzed to determine the reliability of the local motion vector LMV.

Based on the reliability, the weight signal is assigned or ignored to a corresponding local motion vector. Then, the field motion vector determiner 62 multiplies the weighted signals with local motion vectors and adds them to generate a field motion vector.

FIG. 7 illustrates a configuration of the weight generator 61 that adaptively weights the unstable condition of the image data as described above, and FIG. 8 shows the image data of the unstable condition and the correlation diagram COR. The characteristics of the liver are shown. Figure 9a also shows the characteristics between the isolation and isolation weights of the local motion vector LMV, and Figure 9b shows the characteristics between the stability and the stability weights of the local motion vector LMV. It is assumed here that field motion vectors FMV are generated in four local motion detection regions as shown in FIG. 3A.

First, the local motion vectors LMV1-LMV4 received by the weight generator 61 are sequentially generated from the local motion decision unit 24 including Cavg1-Cavg4, Cmax1-Cmax4, and C ₂ nd-C ₂ nd4, which are statistical variables. do. Therefore, the weight generator 61 may store the local motion vectors and statistical variables sequentially generated as described above at the time of generation and process them again at the time of generating the field motion vector FMV. Therefore, the local motion vector LMV stores the local motion vector LMV received every time the MDCK occurs in the registers 751-754. Therefore, as shown in FIG. 3A, when the search of the local motion detection regions is completed and MDCK4 is generated, LMV4 is stored in the register 751, LMV3 is stored in the register 752, and LMV2 is stored in the register 753. 754 stores LMV1, respectively. In the above state, when LMVS1-LMVS0 is generated in the order of 00-01-10-11 as shown in FIG. 3C, the first selector 702 converts the local motion vector of LMV1-LEV2-LMV3-LMV4. Occurs in order. Statistical variables are also generated in the order of Cavg1-Cavg2-Cavg3-Cavg4, Cmax1-Cmax2-Cmax3-Cmax4, and C ₂ nd 1 -C ₂ nd ₂ -C ₂ nd 3 -C ₂ nd ₄ . Therefore, it can be seen that the local motion vectors LMV and statistical variables generated from the same local motion regions are sequentially generated. Here, the local motion vector LMV is used to generate a weight signal, and statistical variables are used as signals for determining the reliability of the corresponding local motion vector LMV.

In case of obtaining the temporary value signal of the local motion vector LMV generated as described above, the weighting signal should be given by analyzing the isolation and stability obtained. First, the isolation means the difference between the local motion vector LMV and the average local motion vector ALMV. Therefore, the subtractor 711 calculates the difference between the local motion vector LMV and the average local motion vector ALMV received from the average local motion vector generator 64, generates an isolation degree, and applies it to the isolation weight coding unit 713. do. Then, the isolation weight coding unit 713 calculates the isolation weight with the same characteristic as that of FIG. 9a according to the isolation received from the subtractor 711. That is, referring to FIG. 9A, the fraud isolation weight has a value of 1 when the isolation degree is 0, and the isolation weight decreases when the isolation degree increases. And if the isolation has a very large value, the isolation weight has a value of zero. Therefore, the isolation weight coding unit 713 analyzes the received isolation degree, and assigns the isolation weight according to the characteristic curve for the isolation degree as shown in FIG. 9a. Second, the stability means the difference between the local motion vector LMV and the previous field motion vector PFMV. Accordingly, the interpolator 712 calculates the difference between the local motion vector LMV and the previous field motion vector PFMV to generate stability, and applies it to the stability weighting coding unit 714. Then, the stability weight coding unit 714 calculates a stability weight with the same characteristics as that of FIG. 9b according to the stability received from the subtractor 712.

That is, referring to FIG. 9b, when the stability is 0, the stability weight has a value of 1. When the stability is increased, the stability weight decreases to approach zero. Therefore, the stability weighting coding unit 714 analyzes the stability received from the subtractor 712, and then assigns the stability weighting value according to the characteristic curve for the stability as shown in FIG. The isolation weights and stability weights generated as described above are added by the adder 715, and the divider 716 divides the weights of the added isolation and stability by 1/2 to generate an averaged weight W. Therefore, the output of the divider 716 is an average value of the isolation weight and the stability weight, and is applied as the weight W of the local motion vector LMV detected in the local motion detection region corresponding to the register 717. The weight signals as described above are sequentially generated by the first selector 702 in the local motion vector LMV, it can be seen that they are sequentially generated according to the corresponding local motion vector LMV.

The conditions for determining the reliability of the image data using the statistical variables will be described. The unstable condition of the image data may be divided into a case having an image having a low contrast, an image having a repetitive image form, a case having a moving object, and the like. First, in the case of an image having a low contrast, the average value of the correlation corresponding to the difference between the two fields decreases, so that the coordinate having the high correlation data COR is increased, so that the accuracy of the detected local motion vector LMV is decreased. Will be. This will be more prominent in the local motion vector generator 11 using the full resolution data after the filter. As described above, the local motion vector generator 11 using the binary edge signal detects the edge by the state of the sub-block due to a predetermined window, so that a satisfactory reliability is obtained even in a low contrast image, but below a certain level. When the contrast is reduced, the edge point is reduced, which also reduces the reliability. Even if the image has an overall contrast on the screen, if the areas that detect local movement are objects having a constant intensity, this area is equivalent to having a low contrast. Therefore, there should be a response when shooting a general subject. . FIG. 8 is a diagram of the characteristics of the topographical degree. When the correlation degree at the position moved by i, f is the highest, the local motion vector LMV is detected by the correlation data COR (i, j). In the case of having (8a), the range of the correlation is large. On the other hand, at low contrast, the signal-to-noise ratio (S / N ratio) of the image signal is reduced, as shown in (8b), so that the correlation average decreases, so the accuracy of the detected moving vector is reduced. In order to adapt to the low contrast as described above, if the average correlation degree Cavg is used as a parameter, the local motion vector LMV obtained in the corresponding region is not used when the threshold value is less than a predetermined threshold. The conditional expression for identifying the image state as described above is the average correlation degree Cavg first threshold value REF1. In the above condition equation, the average correlation degree Cavg is the average number of matched edges, and the first threshold value REF1 can experimentally obtain a satisfactory motion vector at 8 points or more. Therefore, in order to determine whether the image data received as described above has a low contrast, the comparator 723 has a mean phase Cavg output from the divider 57 of the local moving decision unit 24 at one end. The first threshold value REF1 is received and compared with the other terminal. In this case, when the average correlation degree Cavg is greater than the first threshold value REF1, the contrast of the received image data is regarded as normal, and the comparator 723 outputs a normal signal. However, if the average correlation Cavg is less than the first threshold value REF1, the contrast of the received image data is regarded as abnormal, and the comparator 723 outputs an abnormal signal. Second, we look at the case of moving objects in the image data.

Second, we look at the case of moving objects on the screen. In general image data, a moving object exists in the screen. If the moving object is in the local motion detection area, the local motion vector LMV in the local motion detection area is detected differently from the motion caused by hand shaking. Therefore, it is determined whether there is a moving object in the local motion detection area, and there is a moving object, and the local motion vector LMV detected in the corresponding local motion detection area should be ignored. The conditional expression for determining the presence of a moving object in the local motion detection region is the maximum correlation Cmax-average correlation Cavg second threshold REF2. In this case, the mean phase degree Cavg represents a mean value of correlation between all candidate motion vectors in the local motion detection region, and the second threshold value REF2 is a threshold value for determining a moving object. If a moving object occurs in the image signal (8d), it is impossible to accurately match the previous field, so the maximum correlation Cmax is relatively reduced. In order to check whether there is a moving object in the image data as described above, the difference is obtained by subtracting the average correlation degree Cavg from the maximum correlation degree Cmax through a subtractor 721. The comparator 724 inputs the output of the subtractor 721 as one terminal and compares the second threshold value REF2 with the other terminal.

In this case, when the difference between the average correlation Cavg at the maximum correlation degree Cmax is less than the second threshold value REF2, the moving object is regarded as no moving object, and the comparator 724 outputs a normal signal. However, when the difference between the maximum correlation Cmax and the average correlation Cavg is smaller than the second threshold REF2, the moving object is considered to exist in the received image data, and the comparator 724 outputs an abnormal signal. .

Third, the case of having a repetitive image form will be described. In the image data, a certain shape such as a stripe is repeated. When the image having such a repetition form is in the local motion detection region, when the motion is detected by matching the binary edge data of the block unit, the overlapping portion of the pattern is generated, and thus the maximum correlation is also Cmax. The accuracy of the local motion vector LMV detected in the local motion detection region is reduced. Therefore, the local motion vector LMV detected in the repetitive form should be ignored. The conditional expression maximum correlation Cmax-second maximum correlation degree C ₂ nd third threshold value REF3 using the parameter for determining such a state becomes. In the conditional expression, the maximum correlation degree Cmax is the maximum value of the number of etched matches in the local motion generation unit 11, and the second maximum correlation degree C ₂ nd represents the second highest correlation. The third threshold value REF3 is a threshold value for determining a moving object. Even in a low contrast image, the result of the conditional expression as described above is small, so that the repetitive image form and the low contrast can be simultaneously identified. In the repetitive image, as shown in (8c), the difference in correlation becomes smooth. In order to determine the presence or absence of such a repetitive image, the subtractor 722 subtracts the second maximum correlation degree C _2nd from the maximum correlation degree Cmax to generate a difference signal. The comparator 725 receives the output of the subtractor 722 as a comparison input terminal and receives the third threshold value REF3 as a reference input terminal to compare the two signals. In this case, when the difference between the maximum correlation degree Cmax and the second maximum correlation degree C ₂ nd is greater than the third threshold value REF3, the received image data is not regarded as having a repetitive image shape, and the comparator 725 is normal. Output the signal. However, when the difference between the maximum correlation degree Cmax and the second maximum correlation degree C ₂ nd is smaller than the third threshold value REF3, it is assumed that a repetitive image form exists in the image data being used, and the comparator 725 Outputs an abnormal signal.

The reliability of the local motion vector LMV is determined by analyzing the state of the received image data as described above, and the output weight is controlled based on the reliability of the local motion vector LMV. That is, the output of the gate 76 that ORs the outputs of the comparators 73-75 is applied as a control signal of the register 717 that outputs the weight of the corresponding local motion vector LMV, and thus the comparators 723-725. If any one of them outputs an abnormal signal, the register 717 cannot output the generated weight by the AND gate 726. In addition, in the case of a reliable local motion vector LMV, all of the comparators 73-75 output normal signals. In this case, the AND gate 726 outputs normal signals, so that the register 717 corresponds to the corresponding signal. Weights are output to the local motion vector LMV. Therefore, the register 717 outputs or discards the weight of the local motion vector LMV depending on the reliability.

As described above, the three conditional expressions are used to determine the weight of the local motion vector LMV adaptively corresponding to the irregular image environment by the correlation. Therefore, if each local motion detection area is applied and one local motion detection area satisfies this condition, it is excluded from the determination of the field motion vector FMV in consideration of the reliability of the local motion vector LMV obtained from the corresponding local motion detection area. The most basic processing of the field motion vector FMV can be obtained by averaging the reliable local motion vector LMVs. Therefore, the local motion vector LMV with low reliability is ignored and the high reliability field motion vector FMV can be obtained using only selected candidate local motion vectors. However, in the case of averaging reliable local motion vector LMVs, if the localized local motion vector LMV is used by the correlation between the two fields or if the local motion vector LMVs having similar similarities are not used even in the reliable local motion detection region, There is a problem that the accuracy of the motion vector FMV is lost. In order to solve the above problems, the isolation and stability of the detected local motion vector LMV are calculated, and the weight of the local motion vector LMV corresponding to the calculated isolation and stability is determined. In the above case, since the reliable local motion vector LMV is applied, an accurate field motion vector FMV can be determined.

The weights W for the plurality of local motion vectors LMV generated as described above should be averaged and generated as the field motion vector FMV. 10 is a configuration diagram of the field motion vector determiner 62 as described above, and the multiplier 101 sequentially outputs W1 * LMV1, W2 * LMV2, W3 * LMV3, and W4 * LMV4. The adder 102 then sequentially adds the output of the multiplier 101. Therefore, the adder 102 generates W1 * LMV1 + W2 * LMV2 + W3 * LMV3 and W4 * LMV4. In addition, the adder 103 adds and outputs the temporary value signal output from the register 717.

Therefore, the adder 103 outputs W1 + W2 + W3 + W4. The divider 104 then receives the output of the adder 103 as the divisor and receives and divides the output of the adder 102 as the dividend. Therefore, the signal outputting the divider 104 becomes W1 * LMV1 + W2 * LMV2 + W3 * LMV3, W4 * LMV4) / (W1 + W2 + W3 + W4). In this case, if there is a problem in the reliability of the random local motion vector LMV and the random weight signal has a value of 0, the adder 103 does not generate a divisor corresponding to the corresponding local motion vector LMV. Accordingly, it can be seen that the adder 103 generates a detection frequency signal of the local motion vector LV detected in the stable image among the weight signals. Therefore, if there is a local motion vector LMV which is unreliable when the field motion vector FMV is generated, it can be seen that the field motion vector determiner 62 does not calculate the weight of the corresponding local motion vector LMV. The signal output to the divider 104 becomes a field motion vector.

As described above, when the field motion vector FMV is determined by comparing two consecutive fields, the camera user recognizes and corrects the motion even when the subject is photographed for the purpose of intentional panning (centering, tilting). In the case of the correction as described above, the image signal of the screen to be processed is a phenomenon that the image is disconnected whenever the maximum motion vector correction region is exceeded. Therefore, the image must be stabilized to remove the disconnection of the image in consideration of the shaking characteristics of the user. The local motion vector LMV detected by the local motion vector generator 11 represents two-dimensional motion on a continuous field, and the removal of an image motion component by such a motion vector is a field from the first motion direction. By continuously tracking the motion vector FMV, a stable output can be obtained. Therefore, even in intentional panning, a still image appears in the correction region of the motion vector. In order to remove this still image, it is necessary to obtain an image in which the movement continues in the initial direction during the panning operation of the camera. The cumulative form of X (n + 1) = k * X (n) + V (n) has a constant coefficient. This can be done in the form of a linear equation such as In the equation, X (n + 1) is the cumulative motion vector AMV in the current field, X (n) is the cumulative motion vector AMV in the previous field, V (n) is the field motion vector FMV, and k is less than 0. Attenuation coefficient greater than 1 is shown. The attenuation coefficient k is an important factor for determining the amount of motion compensation attenuation of the image and the response characteristics during the panning operation. As the attenuation coefficient k increases, the error rate of the cumulative motion vector AMV required for image stabilization decreases while the response characteristic to panning worsens, resulting in an unnatural disconnection. In addition, if the attenuation coefficient k is small, the cumulative error rate is rapidly increased, and thus the function of image stabilization is lost, but the response to panning is improved. In order to obtain satisfactory attenuation coefficient k in consideration of the characteristics of the trade-off relationship, the error rate relationship between the vibration frequency of the camera and the cumulative motion vector AMV is determined experimentally. In the present invention, the attenuation coefficient is determined to be 0.995 (hereinafter, referred to as a first attenuation coefficient k1) in order to obtain a shake correction characteristic of 95% or more at a hand shake frequency of 2-10 Hz with reference to the experimental result. In addition, in order to improve the panning characteristic, the attenuation coefficient is set to 0.97 or higher (hereinafter referred to as a second attenuation coefficient k2) above the jitter frequency. The attenuation coefficients K1 and K2 are switched by judging whether there is intentional panning and whether a motion vector is corrected. Therefore, the cumulative motion vector AMV converges to zero by the above equation when the camera is stationary. When the first motion vector is generated, the attenuation coefficient is 0.91 and the first attenuation coefficient k1 is set. When the motion vector in the same direction is generated by more than 3-10 frames due to the shaking frequency, the attenuation coefficient is recognized as a panning operation. The second attenuation coefficient k2 is assumed. If the motion vector exists in the correction region, the first attenuation coefficient k1 is set. If the correction area is exceeded, the second attenuation coefficient k2 is set. In addition, the upper limit of the cumulative motion vector AMV, which represents the amount of image stabilization that can be corrected from the first imaging direction, is determined by a digital zooming ratio. To be processed. Therefore, when the upper limit of the cumulative motion vector AMV is exceeded by panning, the reset process is performed. This results in the cutting effect of the image, which is annoying, and thus the limiting process of the cumulative motion vector AMV is performed at the upper limit. .

At this time, when the amount of vibration is in the limit state, the attenuation coefficient should be changed to the first attenuation coefficient k1 of 0.995, which is the image stabilization mode.

FIG. 11 is a configuration diagram of the cumulative motion vector generating unit 13 which compensates for panning by the image stabilization characteristic, FIG. 12a is a timing diagram indicating generation of a panning identification signal PID, and FIG. 12b is a cumulative motion vector AMV and maximum motion. The relationship between the vector correction areas is shown, and FIG. 12C shows a process in which the camera shake is corrected by switching between the attenuation coefficients k1 and k2 in the same state as in FIG. 12B. Looking at the process of suppressing the panning effect according to the configuration, first, the panangang identification unit 111 generates a panning identification signal PID by checking whether the current panning is generated from the received field motion vector FMV and the duct data COR. That is, the panning identifier 111 analyzes the state of the received field motion vector FMV and outputs a panning identification signal PID indicating panning when a motion vector occurs in the same direction over a predetermined frame. The identification signal for selecting the first attenuation coefficient k1 is generated. It is assumed here that the number of frames is 10 frames. The multiplexer 118 that receives the panning identification signal PID generated from the panning identification unit 11 as a selection signal receives the output of the first attenuator 116 as the first terminal and the second attenuator 117 as the second terminal. Receive the output of Here, the first attenuator 116 attenuates the received cumulative motion vector AMV by the first attenuation coefficient k1, and the second attenuator 117 decreases the cumulative motion vector AMV by the second attenuation coefficient k2. Therefore, the multiplexer 118 selects the outputs of the attenuators 116 and 117 according to the state of the panning identification signal PID which is the output of the panning identification unit 111. At this time, when the panning identification signal PID is generated, the multiplexer 118 selects the output of the second attenuator 117 that outputs the cumulative motion vector AMV attenuated by the second attenuation coefficient k2. An output of the first attenuator 116 that outputs the accumulated motion vector AMV attenuated by one attenuation coefficient k1 is selected. The first attenuator 116 attenuates the cumulative motion vector AMV received using a first attenuation coefficient k1 of 0.995 and then applies it to the first terminal of the multiplexer 118, and the second attenuator 117 The cumulative motion vector AMV received using the second attenuation coefficient k2 of 0.97 is attenuated and applied to the second terminal of the multiplexer 118. Then, the multiplexer 118 outputting the panning identifier 111 selects the accumulated motion vector AMV attenuated by a corresponding attenuation coefficient according to the state of the panning identification signal PID and applies it to the adder 113. The adder 113 adds the output of the multiplexer 118 and the field motion vector FMV output from the register 112 to generate the current cumulative motion vector AMV. Thus, the cumulative motion vector generator 13 It can be seen that the field motion vector FMV output from the field motion vector generation unit 12 is accumulated by sequentially adding with the previous cumulative motion vector AMV value. The output of the adder 113 is applied to the limiter 114. When the magnitude of the cumulative motion vector AMV exceeds the upper limit of the correction area, the limiter 114 may accumulate the cumulative motion to prevent the effect of disconnection of the image. Limit the vector AMV. The cumulative motion vector AMV outputting the limiter 114 is output through the register 115.

In addition, the correction region of the cumulative motion vector AMV is determined according to the zoom ratio processed by the address control and zoom processor 14. In other words, when the zoom ratio is large, the correction region of the cumulative motion vector AMV is increased, but the quality of the image signal being zoomed is degraded. Therefore, the proper zoom ratio should be set, and the upper limit of the cumulative motion vector AMV to realize the set zoom ratio should be determined. When the magnitude of the cumulative motion vector AMV reaches an upper limit, it is greatly attenuated so that the cumulative motion vector AMV can be quickly returned to the zero direction. In other words, when the magnitude of the cumulative motion vector AMV value reaches an upper limit, the attenuation coefficient must be reduced to increase the attenuation of the cumulative motion vector AMV. This function is performed by the comparator 119. The comparator 119 receives the cumulative motion vector AMV as a comparison input and receives a fourth threshold value REF4 as a reference input. The fourth threshold value is a threshold value set such that the cumulative motion vector AMV value does not exceed the correction region. The outputs of the comparator 119 and the panning identifier 111 are applied as a selection signal of the multiplexer 118 through the oragate 120. Therefore, the multiplexer 118 performs the same operation as that of generating the panning identification signal PID, and selects the output of the first attenuator 116 when the cumulative motion vector AMV has a size of a normal correction region (AMVREF4) and corrects the output. When the value of the upper limit of the area is exceeded (AMVREF4), the output of the second attenuator 117 is selected. Then, a cumulative motion vector AMV is generated while performing the same operation as the panning identification signal PID.

When the cumulative motion vector AMV is generated as shown in FIG. 12B, when the cumulative motion vector AMV is smaller than the fourth threshold value REF4, the comparator 119 outputs a signal for selecting an output of the first attenuator 116. Thus, the multiplexer 118 selects and applies the cumulative motion vector AMV of the first attenuator 116, which is attenuated to 0.995 having the first attenuation coefficient k1 as shown in FIG. 12c, before the time point T1. . However, after the time point T1, if a motion vector in the same direction is generated at a predetermined frequency by more than 10 frames due to the shaking, the panning identifier 111 generates the panning identification signal PID as shown in FIG. 12a.

Then, the multiplexer 118 selects the output of the second attenuator 117 that is attenuated to 0.97, which is the second attenuation coefficient k2, as shown in FIG. 12c by the panning identification signal PID, and applies it to the adder 113. In this state, when the panning identification signal PID is released at time T2, the multiplexer 118 selects and outputs the output of the first attenuator 116 again. When the output of the first attenuator 116 is selected as described above and the magnitude of the cumulative motion vector AMV becomes greater than the fourth threshold value REF4 as in time T3, the multiplexer 118 again returns the second attenuator 117. Select the output of so that the cumulative motion vector AMV can be quickly reduced to a correctable size. However, as in the time point T4, the cumulative motion vector AMV can be quickly reduced to a correctable size. However, when the cumulative motion vector AMV reaches the upper limit as in the time point T4, the limiter 114 is operated from this time to maintain the cumulative motion vector AMV at a constant size.

Second embodiment

FIG. 13 is a configuration diagram of a second embodiment of an image stabilization system according to the present invention. In the first embodiment of the present invention disclosed in FIG. 13, the output terminal and the address control and zoom processing unit of the cumulative motion vector generator 13 A compensating cumulative motion vector generating unit 16 is connected between the input terminals of 14). Looking at the configuration of FIG. 13 shown as the second embodiment, the digital projection data generated from the camera is applied to the local motion vector generator 11 and the field memory 15. FIG. The local motion vector generator 11 receives digital image data, detects a binary edge signal of the current field from the received image data, and converts the binary edge signal of the detected current field into a binary edge signal of this field. The pattern is matched with the unit of local motion detection area, and the correlation is calculated by comparing two consecutive fields. Using this correlation data, local motion vector LMV and statistical variables are generated in the corresponding local motion detection area. . Receive local motion vector LMVs and statistical variables. The field motion vector generation unit 12 generates a field motion vector FMV representing the total motion of one field from the received fraud local motion vector LMVs and statistical variables. The cumulative motion vector generator 13 is connected to the output terminal of the field motion vector generator 12, and accumulates the received field motion vector to stabilize the shaking between successive fields to an initial state. Occurs. Compensation cumulative motion vector generation unit 16 is connected to the output terminal of the cumulative motion vector generation unit 13, the compensation to detect the false detection movement of the received cumulative motion vector to suppress the instability of the image due to fine noise Generates a cumulative motion vector (Compcensation AMV). The field memory 15 receives and stores the image data, and outputs the image data stored in the corresponding area by the read address. The address control and zoom processor 14 is connected to the output terminal of the compensation accumulation motion vector generator 16, generates a read address from the received compensation accumulation motion vector CAMV, and applies the read address to the field memory 15. Motion compensation is performed by receiving the image data read from the field memory 15. That is, the address control and zoom processing unit 14 performs image compensation received from the field memory 15 by the read address by motion compensation through the above process, and the image signal during the digital zoom process. By interpolating and zooming in on a certain part of the image, the final stable image is output.

In the configuration of the second embodiment, the compensation cumulative motion vector generating unit 16 can reproduce a stable screen by eliminating false detection dominant vector components that may be represented by the cumulative motion vector AMV when freezing. To this end, the compensation cumulative motion vector generating unit 16 uses an adaptive error vector compensation value. I use it. In this case, the smaller the magnitude and maximum correlation difference of the cumulative motion vector AMV, the higher the probability of false detection motion vector. Therefore, the fine motion can be removed by increasing the error vector compensation value. As a result, since the probability of the cumulative motion vector is high, the fine motion can be removed by increasing the error vector compensation value. As a result, when the cumulative motion vector AMV and the average maximum correlation coefficient Cad are large, perfect motion correction can be performed using the detected motion vector to the maximum.

14 is a block diagram of the compensation cumulative motion vector generating unit 16 for suppressing the cumulative motion vector generated as described above including a fine motion vector due to noise, and the minimum motion determining means includes an error vector compensation determining unit ( 141,142). It consists of an adder 143 and a divider 144. The first error vector compensation determiner 141 is connected to the output terminal of the cumulative motion vector generator 13 and determines the minimum motion of the received cumulative motion vector AMV to generate the first error vector compensation value C1. The second error vector compensation determiner 142 is connected to the subtractor 54 and determines the minimum validity of the received average maximum correlation difference Cad to generate the second error vector compensation value C2. The adder 143 is connected to the output terminals of the first error vector compensation determiner 141 and the second error vector compensation determiner 142, and adds two error vector compensation values C1 and C2. The divider 144 is connected to the adder 143 and generates an average error vector compensation value Vc by dividing the output C1 + C2 of the adder 143 by 1/2. Therefore, the minimum motion determining means determines the minimum effective motions of the received cumulative motion vector and the average maximum correlation coefficient Cad, respectively, and averages the two error vector compensation values. The compensation accumulation motion vector generating means is composed of registers 115 and 116, adders 117 and 118, an oar gate 150, an inverter 121 and a multiplexer 17. The register 146 receives and stores the cumulative motion vector AMV.

The register 145 receives and stores the average error vector compensation value Vc. The subtractor 147 receives the outputs of the registers 116 and 115, and subtracts the average minimum motion value Vc from the cumulative motion vector AMV to generate a first compensation accumulation motion vector X (n) -Vc. The adder 148 receives the output of the registers 116 and 115, and adds the cumulative motion vector AMV and the error vector compensation value Vc to generate a second compensation cumulative motion vector X (n) + Vc. The OR gate 150 receives and ORs the MSB of the first compensation accumulation motion vector X (n) -Vc of the subtractor 147 and the MSB of the second compensation accumulation motion vector X (n) + Vc, which is inverted. The multiplexer 149 receives the first compensation accumulation motion vector X (n) -Vc output of the subtractor 147 as a first terminal and the second compensation accumulation motion vector X (n) + Vc output of the adder 148. The terminal receives the second terminal, receives the MSB of the first compensation accumulation motion vector X (n) -Vc as the selection terminal, and receives the output of the gate 120 as the enable terminal.

The multiplexer 149 selects a compensation cumulative motion vector according to the relationship between the cumulative motion vector AMV and the error vector compensation value Vc, and outputs the compensation cumulative motion vector to the address control and zoom processor 14. The final motion vector generating means subtracts the cumulative motion vector and the averaged minimum effective motion value to generate a first compensation cumulative motion vector, and compares the cumulative motion vector with an error vector compensation value Vc to generate a noise component. Compensation cumulative motion vector is generated adaptively to fine movement. That is, when the cumulative motion vector AMV is positive and larger than the error vector compensation value Vc, the first compensation cumulative motion vector X (n) -Vc is output as a compensation cumulative motion vector, and the cumulative motion vector AMV is positive. If the error vector compensation value is less than Vc, the compensation accumulation motion vector is reset to 0. If the cumulative motion vector AMV is negative and the cumulative motion vector AMV is smaller than the error vector compensation value Vc, the second compensation motion vector is calculated. X (n) + Vc is selected and output as a compensation cumulative motion vector, and when the cumulative motion vector AMV is negative and larger than the error vector compensation value Vc, the compensation cumulative motion vector is reset to zero.

FIG. 15A is a characteristic diagram showing the relationship between the cumulative motion vector AMV processed by the first error vector compensation determining unit 141 and the error vector compensation value, and FIG. 15B is processed by the second error vector compensation determining unit 142. FIG. This is a characteristic diagram showing the relationship between the average maximum correlation difference Cad and the error vector compensation value.

Based on the above configuration, the operation of the second embodiment of the image stabilization system according to the present invention will be described. The process of generating the cumulative motion vector AMV is the same as the operation of the first embodiment.

When the image stabilization system is implemented in hardware, experiments show that even a still image signal is slightly shaken by the noise component. This phenomenon occurs because the image signal is converted into binary data by edge detection and processed, so that the noise component affects around the maximum correlation degree Cmax and generates a small amount of motion vector. Movement will occur. In order to eliminate the above problems and provide a stable image in the still picture, the second embodiment suppresses the error motion vector of the cumulative motion vector AMV. This is for retrieving data reduced by the error vector compensation value by inputting the cumulative motion vector AMV and resetting the data corresponding to the low motion vector to zero. Then, the noise component of the motion vector is suppressed to stabilize the reproduction of the screen. However, in such a case, the small actual shake component of the field motion vector FMV is also removed, thereby degrading the function of the image stabilization system. Therefore, the proposed characteristic equations for suppressing the noise components and processing to satisfy the efficiency of the image stabilization system at the same time are as follows.

X (n) is the cumulative motion vector AMV of the current field, Xc (n) is the cumulative motion vector AMV subjected to coring, and Vc represents an error vector compensation value (coring vlaue) indicating that it is the minimum valid motion. In general, the error vector compensation value Vc is an important factor in the performance of signal processing, and the maximum correlation is the difference between the image freeze and the maximum correlation Cmax and the second maximum correlation C ₂ nd due to the magnitude of the cumulative motion vector AMV. The difference Cdif combines the weights of whether the noise component is affected or not so that the range can be adaptively changed. Since the maximum correlation difference Cdif is a signal generated in units of a local motion detection region, the average maximum correlation conductor Cad is obtained by averaging the maximum correlation difference signal by one field period. The average maximum correlation conductor Cad is generated from the average maximum correlation difference generation unit 65. Therefore, if the cumulative motion vector AMV and the average maximum correlation difference Cad are small, the video camera is almost stopped. Therefore, the error vector compensation value is increased to suppress the noise component that can cope with slight movement, and the cumulative motion If the vector AMV and the maximum correlation difference Cad are large, the motion of the video camera is clear. Therefore, the error vector compensation value is reduced to minimize the image stabilization error. Compensation of the cumulative motion vector AMVDPTJ error motion vector was able to obtain a stable image at an error vector compensation value of up to 0.5 pixels. Experimental results show that the coring process is necessary for 5 pixels in the cumulative motion vector AMV, and the average maximum correlation coefficient Cad is required for more than 10 edge matching numbers. In the present invention, as shown in Figs. 15A and 15B, the error vector compensation value preset in accordance with the cumulative motion vector AMV and the average maximum correlation degree Cad is adaptively given. The adaptive coring process for suppressing the noise component of the cumulative motion vector AMV can compensate for the small amount of motion caused by the noise component of the still image while maintaining the accuracy of image stabilization.

The hardware configuration of the cumulative motion vector AMV is shown in FIG. In the present invention, the error belt compensation value is determined by the magnitude of the cumulative motion vector AMV and the average maximum correlation coefficient Cad. That is, when the cumulative motion vector AMV and the normal maximum correlation difference Cad are small, the camera is almost stopped, and the error vector compensation value Vc is increased to suppress the noise. On the other hand, if the cumulative motion vector AMV and the maximum correlation difference Cdif are respectively large, the motion of the camera is clear and the error vector compensation value Vc is reduced to minimize the camera shake correction error.

Accordingly, the first error vector compensation determiner 14 receiving the cumulative motion vector AMV output from the cumulative motion vector generator 13 provides an error vector compensation value of the cumulative motion vector AMV. That is, the first error vector compensation determiner 141 assigns an error vector compensation value according to the magnitude of the cumulative motion vector AMV, as shown in FIG. 12A. When the cumulative motion vector AMV is zero, the maximum value is 0.5, and as the cumulative motion vector AMV is increased, the error vector compensation value decreases linearly to be 0 when the cumulative motion vector AMV is increased. In addition, the second error vector compensation determining unit 142 that receives the average maximum correlation difference Cad output from the subtractor 54 of the local motion decision unit 24 determines the error vector compensation value of the average maximum correlation difference Cad. Grant. That is, the second error vector compensation determiner 142 assigns an error vector compensation value according to the magnitude of the average maximum correlation difference Cad as shown in FIG. 12B. When 0 is 0, the maximum effective correlation decreases as the average maximum correlation difference Cad gradually increases, and when the average maximum correlation difference Cad exceeds 10, it becomes 0. When the error vector compensation values for the cumulative motion vector AMV and the average maximum correlation difference Cad generated as described above are C1 and C2, respectively, the adder 143 and the divider 144 add 2 to C1 and C2, and Divide by to get the average value. The error vector compensation value Vc is obtained by (C1 + C2) / 2 as the average value as the error vector compensation value Vc.

When the error vector compensation value Vc is determined as described above, the motion vector subjected to the fine motion suppression process can be obtained using the characteristic expression described above. That is, when the cumulative motion vector AMV is positive and smaller than the error vector compensation value, and when the cumulative motion vector AMV is negative and larger than the error vector compensation value, the cumulative motion vector AMV is reset to zero without motion compensation. If the cumulative motion vector is positive and larger than the error vector compensation value or if the cumulative motion vector is negative and small, the magnitude of the cumulative motion vector AMV is reduced by the error vector compensation value. In order to suppress the fine movement as described above, the subtractor 147 subtracts the error vector compensation value Vc from the received accumulated motion vector X (n) to generate an X (n) -Vc value, and the adder 148 The error vector compensation value Vc is added to the received accumulated motion vector X (n) to generate an X (n) -Vc value. The outputs of the subtractor 147 and the adder 148 are applied to the first terminal and the second terminal of the multiplexer 149, respectively. The OR gate 150 receives and ORs the sign bit MSB of the subtractor 147 and the inverted sign bitter MSB of the adder 148, and converts the signal generated by the multiplexer 149 into an enable signal. Is authorized. Therefore, when the sign bit of the subtractor 147 is 1 and the sign bit of the adder 148 is 0, that is, the cumulative motion vector AMV is positive and the magnitude is smaller than the error vector compensation value, or the cumulative motion vector AMV is negative. If the error vector compensation value Vc is larger than the error vector compensation value, the OR gate 150 outputs a logic high signal. In this case, the multiplexer 149 is disabled to output the final cumulative motion vector Xc (n) as zero. . Therefore, when the cumulative motion vector AMV has a low motion vector component, the cumulative motion vector AMV is reset to suppress the fine motion vector component represented by the noise component. However, except for the above conditions, the gate 120 enables the multiplexer 149. At this time, the sign bits of the first compensation accumulation motion vector X (n) -Vc outputting the subtractor 147 are applied as the selection signal of the multiplexer 149. Accordingly, according to the output logic of the subtractor 147, the multiplexer 149 receives the first final motion vector X (n) when the cumulative motion vector X (n) is positive and larger than the error vector compensation value Vc. A second cumulative motion vector, which is output as a final cumulative motion vector X (n) that selects a Vc value and is received as a second terminal when the cumulative motion vector X (n) is negative and smaller than the error vector compensation value Vc. Select X (n) + Vc and output it as the final cumulative motion vector Xc (n). Therefore, when the camera motion is clear, the motion blur vector AMV is reduced by the error vector compensation value Vc to suppress the fine motion vector component caused by the noise.

Therefore, as described above, the fine motion components included in the cumulative motion vector AMV are removed by the coring processor. Thus, the cumulative motion vector generator 13 receives a reliable motion vector component signal from the address control and zoom processor. (14) can be supplied. Therefore, the address control and zoom processor 14 can precisely control the access switch of the image data stored in the field memory 15 by the cumulative motion vector AMV. Therefore, the address control and zoom processing unit 14 can access a stable zoom image by enlarging a predetermined portion of the image by interpolation of the image data stored in the field memory 15.

Third embodiment

16 is a configuration diagram of a third embodiment of an image stabilization system according to the present invention. The third embodiment of the present invention disclosed in FIG. 16 is the field motion vector generator 12 in the first embodiment of FIG. The compensating field motion vector generator 17 is connected between the output terminal of the output terminal and the input terminal of the cumulative motion vector generator 13. Referring to the configuration of FIG. 16 shown in the third embodiment, the digital image data generated from the camera is applied to the local moving vector generator 11 and the field memory 15. The local motion vector generator 11 receives digital image data, detects a binary edge signal of the current field from the received image data, and converts the detected binary edge signal of the current field into a binary edge signal of the previous field. The pattern is matched by using the unit and the local motion detection area unit to calculate the correlation by comparing two consecutive fields.The correlation data is used to generate the local motion vector LMV and statistic variables in the corresponding local motion detection area. do. The field motion vector generator 12 is connected to the output terminal of the local motion vector generator 11 to receive local motion vector LMVs and statistical variables.

The field motion vector generation unit 12 generates a field motion vector FMV representing the overall motion of one field according to the stability of the image from the received local motion vector LMVs and statistical variables. The compensation field motion vector generator 17 is connected to the output terminal of the field motion vector generator 12 to receive the field motion vector FMV and the previous cumulative motion vector PAMV. The compensation field motion vector generation unit 17 receives the previous cumulative motion vector PAMV to determine the magnitude of the fine motion, and removes the error motion vector included in the received field motion vector FMV according to the magnitude of the fine motion. To generate a Compensating Field Motion Vector (CFMV). The cumulative motion vector generator 13 is connected to the output terminal of the compensation field motion vector generator 17 and receives the compensation field motion vector CFMV. The cumulative motion vector generating unit 13 accumulates the received compensation field motion vector CFMV and generates a cumulative motion vector AMV for stabilizing the shaking between successive fields to an initial state, and the compensation field motion vector generating unit It is applied by input of (17). The field memory 15 receives and stores the image data, and outputs the image data stored in the corresponding area by the read address. The address control and zoom processor 14 is connected to the output terminal of the compensation field motion vector generator 16, generates a read address from the received accumulated motion vector AMV, and applies the read address to the field memory 15. Motion compensation is performed by receiving image data read from the memory 15. That is, the address control and zoom processor 14 performs image compensation received from the field memory 150 by the read address by motion compensation through the above process, and interpolates the image signal during the digital zoom process. (interploation) to enlarge a certain portion of the image and finally output the stabilized image.

In the configuration of the third embodiment, the compensation field motion vector generator 17 removes an error motion vector included in the received field motion vector FMV, thereby removing a fine motion component close to a still picture from the field motion vector FMV. It becomes possible to reproduce a stable image. Here, the error vector compensation value for adaptively coring the field motion vector FMV may be implemented in the same form as in the second embodiment.

That is, the compensation field motion vector generation unit 17 uses an adaptive error vector compensation value, and uses the previous cumulative motion vector PAMV and the average maximum correlation coefficient Cad as elements for determining the error vector compensation value. In this case, the smaller the size of the previous cumulative motion vector PAMV and the smaller the maximum correlation difference Cdif, the higher the probability of misdetection motion vector. have. As a result, when the cumulative motion vector PAMV value is large or the average maximum correlation coefficient Cad is large, perfect motion correction can be performed using the detected motion vector to the maximum.

17 is a configuration diagram of a compensation field motion vector 17 for suppressing the field motion vector CFMV including a fine motion vector due to noise. 143 and divider 144.

The first error vector compensation determiner 141 is connected to the output terminal of the cumulative motion vector generator 13 and determines the minimum motion of the previous cumulative motion vector PAMV received through the field delay 152 to compensate the first error vector. Generates the value C1. The second error vector compensation determiner 142 is connected to the subtractor 54 and determines the minimum movement of the received maximum correlation difference Cdif to generate the second error vector compensation value C2. The adder 143 is connected to the output terminals of the first error vector compensation determiner 141 and the second error vector compensation determiner 142 and adds C1 and C2. The divider 144 is connected to the fraud adder 143 and divides the output C1 + C2 of the adder 143 by 1/2 to generate Vc, which is an average error motion vector compensation value. Therefore, the minimum motion determining means determines the error vector compensation values of the previous cumulative motion vector PAMV and the average maximum correlation coefficient Cad respectively, and averages the two error vector compensation values and outputs the error vector compensation value Vc. The compensation motion vector generating means is composed of registers 145 and 146, a subtractor 147, an adder 148, an oar gate 150, an inverter 151 and a multiplexer 149. The register 146 receives and stores the field motion vector FMV. The resistor 145 receives, stores, and stores a fraud error vector compensation value Vc. The subtractor 147 receives the outputs of the registers 146 and 145, and subtracts the error vector compensation value Vc from the field motion vector FMV to generate a first compensation field motion vector X (n) -Vc. The adder 148 receives the output of the registers 116 and 115, and adds the field motion vector FMV and the error vector compensation value Vc to generate a second compensation field motion vector X (n) + Vc. The oragate 150 receives and outputs the MSB of the second compensation field motion vector X (n) + Vc inverted from the MSB of the first compensation field motion vector X (n) + Vc. The multiplexer 149 receives the first compensation field motion vector X (n) + Vc output of the subtractor 147 as a first terminal and outputs the second compensation field motion vector X (n) + Vc of the adder 148. The terminal receives the second terminal, receives the MSB of the first compensation field motion vector X (n) -Vc as the selection terminal, and receives the output of the gate 120 as the enable terminal. The multiplexer 149 generates a compensation field motion vector CFMV according to the relationship between the field motion vector FMV and the error vector compensation value Vc, and outputs it to the address control and zoom processor 140. The compensation motion vector generating means is the field motion vector A first compensation field motion vector is generated by subtracting an error vector compensation value Vc, which means a vector FMV and an averaged minimum motion value, and when the field motion vector FMV is positive and larger than the error vector compensation value Vc, The compensation field motion vector X (n) -Vc is output as the compensation field motion vector CFMV, and when the field motion vector FMV is negative and smaller than the error vector compensation value Vc, the second compensation field motion vector X (n). Select + Vc as the compensation field motion vector CFMV, and the field motion vector FMV is positive and smaller than the error vector compensation value Vc or the fill. While motion vector FMV is negative, the magnitude is greater than the error vector compensation value Vc to reset the compensation field motion vector CFMV 0 will suppress noise components.

FIG. 15A is a characteristic diagram showing the relationship between the previous cumulative motion vector PAMV processed by the first error vector qhtkd determining unit 141 and the error vector compensation value, and FIG. 15B is processed by the correlation difference motion determining unit 112. FIG. This is a characteristic diagram showing the relationship between the average maximum correlation difference Cad and the error vector compensation value.

FIG. 18 is a configuration diagram of the cumulative motion vector generating unit 13 of FIG. 16, and the panning identification unit 111 receives the compensation field motion vector CFMV and the correlation data COR, and the compensation field motion vector CFMV is equal to or greater than a predetermined frame. When generated in the same direction, it is considered as deliberate panning and generates a panning identification signal PID for changing the attenuation coefficient of the cumulative motion vector AMV. In addition, the comparator 119 receives a cumulative motion vector AMV as a comparison input, and receives a fourth threshold value REF4 as a reference input so that the magnitude of the cumulative motion vector AMV does not exceed the maximum correction region. The comparator 119 generates a comparison result signal for changing the attenuation coefficient of the cumulative motion vector AMV when the magnitude of the cumulative motion vector AMV is larger than the fourth threshold value REF4. OA gate 120 receives the output of the pan identification unit 111 and the comparator 119, and applies the received signal as a selection signal of the attenuation coefficient. The attenuation determining means is composed of a multiplexer 118 and attenuators 116 and 117. The first attenuator 116 receives the cumulative motion vector AMV up to the previous time, and attenuates the cumulative motion vector AMV received with the set first attenuation coefficient K1. The second attenuator 117 receives the cumulative motion vector AMV up to the previous, and attenuates the cumulative motion vector AMV received with the second attenuation coefficient K2. The multiplexer 118 receives the output of the first attenuator 116 as a first terminal, receives the output of the second attenuator 117 as a second terminal, and outputs the output of the oragate 120 to a selection terminal. Receive. The multiplexer 118 selects and outputs the output of the second attenuator 117 when the panning identification signal PID or the comparison result signal is received. Otherwise, the multiplexer 118 selects and outputs the output of the first attenuator 116. . The cumulative motion vector generating means is composed of registers 92 and 95, an adder 113, and a limiter 114. The register 112 stores the received compensation field motion vector CFMV. The adder 113 receives the outputs of the register 112 and the multiplexer 118 and adds the two signals to generate a cumulative motion vector AMV. The limiter 114 receives the output of the adder 113 and limits when the cumulative motion vector AMV is more than a predetermined size. The register 115 receives and stores the output of the limiter 114, applies it to the inputs of the attenuators 116, 117 and the comparator 119, and outputs it to the address control and zoom processor 14.

FIG. 12A shows a shape signal of the panning identification signal PID, and FIG. 12B shows an example of the movement amount of the cumulative motion vector AMV according to the change in time, and FIG. 12C shows the 12B when panning is detected from the field motion vector FMV. As shown in the figure, the attenuation coefficient is switched to generate the cumulative motion vector AMV.

Based on the above configuration, the operation of the third embodiment of the image stabilization system according to the present invention will be described. The process of generating the field motion vector FMV is the same as the operation process of the first embodiment.

As described above, when the image stabilization system is implemented in hardware, slight shaking is detected by the noise component in the still image signal. This phenomenon occurs because the image signal is converted into binary data by edge detection and processed, so that the noise component affects around the maximum correlation Cmax and generates a small amount of motion vector, which is included in the field motion vector FMV. Because. Therefore, when the cumulative motion vector generating unit 13 accumulates the field motion vector FMV and generates a cumulative motion vector AMV, such a misdetected motion vector is accumulated to generate unnecessary continuous motion over time. In order to eliminate the above problems and provide a stable image in the still image, the third embodiment suppresses the error motion vector included in the field motion vector FMV. This is to reset the data corresponding to the fine motion vector to zero by outputting the data reduced by the error vector compensation value by inputting the field motion vector FMV including the fine misdetection motion vector as described above. Then, the noise component of the motion vector is suppressed to stabilize the reproduction of the screen. However, in such a case, the small actual shaking component of the field motion vector FMV is also removed, thereby degrading the function of the image stabilization system. Therefore, the proposed characteristic equations for suppressing noise components and processing to satisfy the efficiency of the image stabilization system at the same time are as follows.

X (n) is the current field motion vector FMV, Xc (n) is the corrugated field motion vector CFMV, and Vc represents an error vector compensation value indicating a minimum valid motion. In general, the error vector compensation value Vc is an important factor in the performance of the signal processing, whether the stop of the image by the size of the vector PAMV previously accumulated motion and maximum correlation Cmax and the second maximum correlation, the largest correlation difference of C ₂ nd The range can be adaptively changed by combining the weights of the influences of noise components by the one-field average signal of the on-chip Cdif.

That is, when the previous cumulative motion vector PAMV and the average maximum correlation difference Cad are small, the video camera is almost stopped. Therefore, the error vector compensation value is increased to suppress the noise component corresponding to the slight movement. If the cumulative motion vector PAMV and the average maximum correlation difference Cad are respectively large, the motion of the video camera is insignificant, so the error vector compensation value is reduced to minimize the image stabilization error. A configuration for removing fine noise of the field motion vector FMV is shown in FIG. As described above, when the error motion vector included in the field motion vector FMV is suppressed, a stable image signal is obtained. In the third embodiment, in order to adaptively suppress the error motion vector of the field motion vector FMV, the error vector compensation value is determined by the magnitude of the previous cumulative motion vector PAMV and the average maximum correlation coefficient Cad, as in the second embodiment. do. That is, when the previous cumulative motion vector PAMV and the average maximum correlation coefficient Cad are respectively small, the camera is almost stopped. Therefore, the error vector compensation value Vc is increased to suppress noise. On the other hand, if the previous cumulative motion vector PAMV and the average maximum correlation coefficient Cad are respectively large, the camera motion is clear and the error vector compensation value vc is reduced to minimize the camera shake correction error.

Accordingly, the first error vector compensation determiner 141 receiving the previous cumulative motion vector PAMV provides an error vector compensation value indicating the minimum effective motion of the previous cumulative motion vector PAMV. That is, the first error vector compensation determiner 141 assigns an error vector compensation value according to the size of the previous cumulative motion vector PAMV, as shown in FIG. 12A. In addition, the second error vector compensation determining unit 142, which receives the average maximum correlation difference Cad output from the subtractor 54 of the local motion decision unit 24, indicates the minimum effective movement of the average maximum correlation difference Cad. Give the error vector compensation value. That is, the second error vector compensation determiner 142 assigns an error vector compensation value according to the magnitude of the average maximum correlation difference Cad as shown in FIG. 12B. When the difference Cad is 0, the maximum 0.5 is obtained. As the mean maximum correlation difference Cad increases gradually, the minimum effective movement decreases, and when the mean maximum correlation difference Cad exceeds 10, it becomes 0. If the error vector compensation values for the previous cumulative motion vector PAMV and the average maximum correlation coefficient Cad generated as described above are C1 and C2, respectively, the adder 143 and the divider 144 add the C1 and C2 and then add them. Divide by 2 to get the average value. The error vector compensation value Vc is obtained by (C1 + C2) / 2 as the average value as the error vector compensation value Vc.

When the error vector compensation value Vc is determined as described above, the compensation field motion vector CFMV subjected to the fine motion suppression process can be obtained using the above characteristic equation. In order to suppress fine movement as described above, the subtractor 147 subtracts the error vector compensation value Vc from the received field motion vector X (n) to generate an X (n) -Vc value, and the adder 148 The error vector compensation value Vc is added to the received field motion vector X (n) to generate a value of X (n) + Vc.

The outputs of the subtractor 147 and the adder 148 are applied to the first terminal and the second terminal of the multiplexer 149, respectively. The OR gate 150 receives and ORs the sign bit MSB of the subtractor 147 and the inverted sign bit MSB of the adder 148, and enables the signal generated by the multiplexer 149. Applied by signal. Therefore, when the sign bit of the subtractor 147 is 1 and the sign bit of the adder 148 is 0, that is, the field motion vector FMV is positive and the magnitude is smaller than the error vector compensation value Vc or the field motion vector FMV is If negative and larger than the error vector compensation value Vc, the oragate 150 outputs a logic high signal. In this case, the multiplexer 149 is disabled to output the compensation field motion vector Xc (n) as 0. . Therefore, when the received field motion vector FMV has a low motion vector component, the compensation field motion vector CFMV is reset to suppress fine noise components. However, except for the above state, the gate 120 enables the multiplexer 149. At this time, the sign bits of the first compensation field motion vector X (n) -Vc outputting the subtractor 147 are applied as the selection signal of the multiplexer 149. Therefore, according to the output logic of the subtractor 147, the multiplexer 149 receives the first terminal motion vector X received by the first terminal when the received field motion vector X (n) is greater than the error vector compensation value Vc. The second compensation field motion vector X (n) is outputted as a compensation field motion vector CFMV, and when the field motion vector X (n) is negative and smaller than the error vector compensation value Vc, the second compensation field motion vector X (n) is output. Select) + Vc to output the compensation field mover CFMV. Therefore, the resulting compensation field motion vector CFMV outputs a value reduced by an error vector compensation value, thereby suppressing a fine motion vector of noise components included in the field motion vector FMV.

Then, the cumulative motion vector generator 13 as shown in FIG. 17 receives the compensation field motion vector CFMV generated as described above to generate the previous cumulative motion vector PAMV. The compensation field motion vector CFMV is a motion vector component from which an erroneously detected motion vector that may be included in the field motion vector FMV is removed. Therefore, since the minute noise component included in the field motion vector FMV is suppressed, the cumulative motion vector generator 13 can generate the cumulative motion vector AMV with stable operation. The operation of the cumulative motion vector generator 13 is performed in the same manner as in the above-described first and second embodiments.

Therefore, the fine motion microcomponents included in the field motion vector FMV are removed by the coring processor. Thus, the cumulative motion vector AMV can supply a reliable motion vector component signal to the address control and zoom processor 14. It becomes possible. Therefore, the address control and zoom processor 14 can accurately control the access switch of the image data stored in the field memory 15 by the cumulative motion vector AMV. Therefore, the address control and zoom processing unit 14 can access a stable zoom image by enlarging a predetermined portion of the image by interpolation of the image data stored in the field memory 15.

As described above, the image stabilization system of the present invention adaptively processes the irregular phenomenon of the image to determine the correct motion vector. The adaptive image stabilization system as described above can be applied to a video camera or a VCR. In the image stabilization system as described above, there are switching of attenuation coefficients based on motion and panning motion identification, control of image characteristics and adaptive system control based on weights, and noise removal of motion vectors by determination of still and hand shake conditions. . In addition, since the adaptive image stabilization system can be implemented in pure hardware, the operation time can be shortened, and the system is easily integrated.

Claims (33)

An image stabilization apparatus comprising a memory for storing image data in field units and a controller for correcting a movement of the image data stored in the memory based on the received cumulative motion vector, wherein the image stabilization device receives the image data and receives the received image. By changing the data into binary edge data, matching the pattern of the edge data of the current field and the previous field, the local motion detection area units sequentially generate the correlation data, and analyze the generated correlation data. Means for sequentially generating statistical variables and local motion vectors of the local motion detection regions, receiving the local motion vectors and statistical variables, and calculating an isolation degree of the local motion vector to correspond to a preset ratio; The isolated weights are adaptively generated and sequentially generated Means for generating a field motion vector by multiplying the lip weight values and the corresponding local motion vectors and averaging them by field periods, and having attenuating means and receiving the field motion vector, and the previous cumulative motion vector attenuated by the attenuation means. And means for generating the cumulative motion vector by accumulating the received field motion vector. The apparatus of claim 1, wherein the means for generating the local motion vector comprises: edge detection means for receiving the image data, analyzing the received image data, and converting the field image into binary edge data; and at least two local movements. A storage means for receiving the binary edge data, sequentially extracting block data of the reference region positions of the local motion detection regions, delaying one field, and outputting the edge detection means. Receive binary edge data as comparison data and receive binary; data of previous field outputting the memory means as reference data, pattern matching of block data of two block data, and correlating data of corresponding local motion detection areas. An edge pattern matching means that is sequentially generated and the correlation data, and receives a corresponding local motion detection region A local motion corresponding to the position value of the maximum correlation diagram while generating the state of the correlation data which is changed by comparing the correlation data received at the corresponding local motion detection region with the previous statistical variables detected as a statistical variable. An image stabilization device comprising local motion determining means generated by a local motion vector of a detection region. 3. The comparison means according to claim 2, wherein the local motion determining means generates a first comparison signal when the correlation degree data is large compared with a previous maximum correlation of the local motion detection area corresponding to the received correlation data. Receiving an address clock and an output of the comparison means, counting the address clock to generate a position value of the received correlation diagram, and corresponding to the position value of the received correlation diagram data when the first comparison signal is received. Receives local motion vector detection means generated as a local motion vector of a local motion detection region, and outputs of the received correlation data and comparison means, and generates the received correlation data when the first comparison signal is received. Means for receiving the output of the maximum correlation and generating means and the comparison means, and the previous maximum correlation when receiving the first comparison signal. Means for generating a second maximum correlation, means for receiving the maximum correlation and the second maximum correlation, subtracting the two received signals to generate a maximum correlation, and receiving the received correlation data. And means for generating the statistical variables of corresponding local motion detection regions by means of generating correlation coefficient data and the added correlation diagram data by the number of candidate motion vectors and generating the average correlation. 4. The method of claim 3, wherein the means for generating the field motion vector receives the statistical variables sequentially output from the local motion determining means, and compares the statistical variables with predetermined thresholds to determine the reliability of the local motion vector. Reliability determination means for judging the occurrence of an abnormal condition that reduces the error and generating a non-base signal when an abnormal condition occurs, and the local motion vector and the average local motion vector generated in units of fields sequentially output from the local motion vector determination means. Calculate the isolation degree of the corresponding local motion vector by subtracting the motion vector information, and generate the adaptive weighting signal at a preset ratio according to the calculated result, and the reliability judging means Means for controlling the output of an isolated weighted signal; And a mean weight calculation means for sequentially receiving local motion vectors, and sequentially applying the weight signals to the local motion vectors of the received local motion detection regions, and averaging them to generate a field motion vector. Video stabilization device. 5. The first comparison means according to claim 4, wherein the reliability judging means compares the average correlation and the first threshold value, which is a reference value of low contrast, and generates an abnormal signal when the average correlation is less than the first threshold value. and; The difference signal is generated by subtracting the maximum correlation value and the average correlation value, and the difference signal is compared with a second threshold value which is a reference value for determining a moving object. Generating second comparing means; Another signal is generated by subtracting the maximum correlation and the second maximum correlation, and the reference value for determining the difference signal and the repetitive image compares the second threshold value, and the difference signal is smaller than the third threshold value. And a third comparing means for generating an abnormal signal at the time of occurrence. 6. The method of claim 5, wherein the weight signal generating means receives the local motion vector and the previous field motion vector, calculates a difference by calculating a difference between two motion vectors, and calculates the stability at a preset ratio according to the calculated stability. A means for generating a weighted value when the stability is small and a weighted value when the stability is small, and adaptively generating a stability weighted signal; And means for generating a weighted signal of a corresponding local motion vector by averaging the received two weights. 7. The apparatus according to claim 4 or 6, wherein the field motion determining means generates the weighted local motion vectors by multiplying the sequentially received local motion vectors with corresponding weight signals, respectively, and the weight signal. And means for adding the weighted local motion vectors, and means for generating a field motion vector by dividing the added local motion vector by the added weight signal. 5. The apparatus of claim 4, wherein the means for generating the cumulative motion vector comprises: means for inputting the previous cumulative motion vector to attenuate it into a first attenuation coefficient; and inputting the previous cumulative motion vector to the first attenuation coefficient; . Means for attenuating with a second attenuation coefficient having a smaller value, and receiving the field motion vector, analyzing the field motion vector, and panning identification for generating a panning signal by considering the motion vector in the same direction over a predetermined frame. Means; Receives the outputs of the first attenuation means and the second attenuation means and receives the output of the panning identification means as a selection signal, selects the output of the second attenuation means upon receipt of the panning signal, or else the first attenuation means. And means for selecting an output of the means, means for outputting the field motion vector and the selection means, and generating means for generating a cumulative motion vector by adding the two signals and outputting the accumulated motion vector to the controller. . 9. The apparatus of claim 8, further comprising: means for receiving a threshold value, which is an upper limit value of the cumulative motion vector and the cumulative motion vector, and generating a second attenuating means selection signal of the selection means when the magnitude of the cumulative motion vector is larger than the threshold value; And a limiter connected between the adding means and the control unit to maintain the cumulative motion vector at a constant value when the cumulative motion vector exceeds a maximum correction region of the motion vector. An image stabilization apparatus comprising a memory for storing image data in field units and a controller for correcting movement of image data stored in the memory by a received compensation accumulation motion vector, the image stabilizing apparatus receiving and receiving image data. Convert the image data to binary edge data, match the pattern of the edge data of the current field and the previous field, and generate the correlation data of the local motion detection area units sequentially, and analyze the generated correlation data Means for sequentially generating statistical variables and local motion vectors of the local motion detection regions, and storing the local motion vectors and statistical variables sequentially and sequentially, and sequentially storing the isolation and stability of the stored local motion vectors. Calculate the weighted signals corresponding to the preset ratio And a means for generating a field motion vector by mixing the weights and the corresponding locally generated weights which are adaptively generated and sequentially generated, and averaging them in a field period, and attenuating means and receiving the field motion vector. Means for generating the cumulative motion vector by accumulating the previous cumulative motion vector and the received field motion vector outputting the attenuation means, receiving the cumulative motion vector, and reducing the cumulative motion vector by error motion compensation. And a compensation cumulative motion vector generating means for generating a compensation cumulative motion vector in which the detected motion vector is suppressed. 11. The apparatus of claim 10, wherein the means for generating a local motion vector comprises: edge detection means for receiving the image data, analyzing the received image data, and converting the field image into binary edge data; and at least two local motions. A storage means for receiving the binary edge data, sequentially extracting blockdales of the reference region positions of the local motion detection regions, and delaying one field, and outputting the edge detection means. Receive binary edge data as comparison data, receive binary edge data of the previous field outputting the memory means as reference data, pattern-match the block data of two block data, and correlate the correlation diagram data of corresponding local motion detection areas. An edge pattern matching means that is sequentially generated and the correlation data, and receives a corresponding local motion detection region Comparing the correlation data received from the previous statistical variables detected in the corresponding local motion detection area, the state of the correlation data that is changed as a statistical variable, while generating a local correlation corresponding to the position value of the maximum correlation An image stabilization apparatus comprising a local motion determining means generated by a local motion vector of a detection permanent. 12. The apparatus according to claim 11, wherein the local motion determining means generates a first comparison signal when the correlation degree data is large compared with a previous maximum correlation of the local motion detection area corresponding to the received correlation data. Receiving an address clock and an output of the comparison means, counting the address clock to generate a position value of the received correlation diagram, and corresponding to the position value of the received correlation diagram data when the first comparison signal is received. Receives local motion vector detection means generated as a local motion vector of a local motion detection region, and outputs of the received correlation data and comparison means, and generates the received correlation data when the first comparison signal is received. And a maximum correlation degree before receiving the first comparison signal and receiving outputs of the maximum correlation degree generating means and the comparison means. Means for generating a second maximum correlation, means for receiving the maximum correlation and the second maximum correlation, subtracting the two received signals to generate a maximum correlation, and the received correlation data previously correlated And means for generating the statistical variables of corresponding local motion detection regions by means of generating mean correlation by dividing the added data with the added data by the number of candidate motion vectors. The method of claim 12, wherein the means for generating the field motion vector receives the statistical variables sequentially output from the local motion determining means, and compares the statistical variables with predetermined thresholds to determine the reliability of the local motion vector. Reliability determination means for judging the occurrence of abnormal conditions that reduce the error and generating an abnormal signal when the non-imaginary condition occurs, the local motion vector and the local local motion vector generated in units of fields sequentially output from the local motion determination means and Receive previous field motion vectors, subtract the motion vector information to calculate the isolation and stability of the corresponding local motion vector, generate the adaptive weighting signal at a preset ratio according to the calculated result, and the reliability The number of controlling the output of the weight signal in accordance with the output of the determination means And sequentially receiving the weighted signals and the local motion vectors, and sequentially applying the weighted signals to the local motion vectors of the received local motion detection regions, and averaging the weighted signals to generate a field motion vector. Image stabilization apparatus comprising a means. The first comparison means according to claim 13, wherein the reliability judging means compares the average correlation with a first threshold value which is a reference value of low contrast, and generates an abnormal signal when the average correlation is less than the first threshold value. And a first comparing means for generating an abnormal signal when the maximum correlation and the average correlation are less than a first threshold value, and generating a difference signal by subtracting the maximum correlation and the average correlation. Second comparing means for comparing a second threshold value, which is a reference value for determining a difference, and generating an abnormal signal when the difference signal is less than the second threshold value; The difference signal is generated by subtracting the maximum correlation value and the second maximum correlation value, and the difference signal is compared with a third threshold value, which is a reference value for determining a repetitive image, and the difference signal is smaller than the third threshold value. And a third comparison means for generating an abnormal signal at the time of occurrence. 15. The method of claim 14, wherein the weight signal generating means receives the local motion vector and the average local motion vector, calculates the degree of isolation by calculating the difference between the two motion vectors, and sets a preset value according to the calculated degree of isolation. A means for adaptively generating an isolated weight value signal by applying a weight when the isolation is large and a weight when the isolation is small, and receiving the local motion vector and the previous field motion vector. Stability is calculated by calculating the difference of the motion vectors, and when the stability is large, the weight is small and the weight is large when the stability is small according to the calculated stability. Means, receiving the isolated weight value signal and the stability weight value signal, and controlling the output of the telemetric determination means. Arc received, the image stabilizing apparatus further comprise means for generating a local motion vector of the corresponding weight signal by averaging the two weights is received. 16. The apparatus of claim 15, wherein the field motion determining means comprises: means for generating weighted local motion vectors by multiplying weight signals corresponding to the sequentially received local motion vectors, and means for adding the weight signals; And means for adding the weighted local motion vectors and dividing the added local motion vector by the added weight signal to generate a field motion vector. 17. The method of claim 16, wherein the means for generating the cumulative motion vector comprises: means for inputting the previous cumulative motion vector to attenuate it into a first attenuation coefficient, and inputting the previous cumulative motion vector to the first attenuation coefficient; Means for attenuating with a second attenuation coefficient having a smaller value, and receiving the field motion vector, analyzing the field motion vector, and panning identification for generating a panning signal by considering the motion vector in the same direction over a predetermined frame. Means for receiving the outputs of the first attenuating means and the second attenuating means and receiving the outputs of the panning identification means as a selection signal, and selecting an output of the second attenuating means upon receiving the panning signal; Means for selecting an output of a first attenuating means, receiving the field motion vector and the output of the selecting means, and adding the two signals to accumulate the motion vector; And a means for generating a data output to the controller. 18. The method of claim 17, wherein the means for generating a compensation field motion vector receives the cumulative motion vector, examines the magnitude of the cumulative motion vector, and determines an error motion compensation phase value when the cumulative motion vector has a small size at a predetermined ratio according to the magnitude. Means for adaptively generating an error motion compensation value by decreasing an error motion compensation value when it is large and having a large value, receiving the error motion compensation value and a baby cumulative motion vector, and receiving the error motion in the cumulative motion vector. Means for generating a first compensation cumulative motion vector by subtracting a compensation value, receiving the error motion compensation value and the cumulative motion vector, and adding the cumulative motion vector and the error motion compensation value to add a second compensation cumulative motion vector. Means for generating, comparing the error motion compensation value with the magnitude of the cumulative motion vector, Is positive and greater than the error motion compensation value, the first compensation cumulative motion vector is selectively outputted. If the cumulative motion vector is positive and less than the error motion compensation value, the compensation cumulative motion vector value is reset and the cumulative motion vector is reset. Is negative and less than the error motion compensation value, the second compensation cumulative motion vector is selectively outputted; and if the cumulative motion vector is negative and greater than the error motion compensation value, the compensation cumulative motion vector is reset. Image stabilization device. 18. The error motion compensation value of claim 17, wherein the means for generating a compensation cumulative motion vector receives the cumulative motion vector, examines the magnitude of the cumulative motion vector, and measures an error motion compensation value when the cumulative motion vector has a small magnitude at a predetermined ratio according to the magnitude. Means for adaptively generating a first error motion compensation value by decreasing the error motion compensation value when it is large and having a large value, and receiving the maximum correlation differences generated from each of the local motion detection areas, and then averaging one field. Means for generating a maximum correlation difference, and checking the magnitude of the average maximum correlation difference and increasing the error motion compensation value when having a small size at a predetermined ratio according to the size and compensating for the error motion when having a large value Means for adaptively generating a second error motion compensation value by reducing a value, the first error motion compensation value and the second error motion compensation value; Means for generating an error motion compensation value by averaging a value, receiving the averaged error motion compensation value and the cumulative motion vector, and subtracting the error motion compensation value from the cumulative motion vector to obtain a first compensation cumulative motion vector. Means for generating, means for receiving the averaged error motion compensation value and the cumulative motion vector, adding the cumulative motion vector and the error motion compensation value to generate a second compensation cumulative motion vector, and the error motion compensation Compare the value with the magnitude of the cumulative motion vector, and if the cumulative motion vector is positive and greater than the error motion compensation value, output the first compensation cumulative motion vector, and the cumulative motion vector is positive and smaller than the error motion compensation value. If the cumulative motion vector is negative and the cumulative motion vector is negative Is smaller than the value of the second selected compensation accumulated motion vector output, image stabilizer, characterized in that consists of means for the accumulated motion vector is negative and vector the reset is larger and the compensation accumulated motion than the error motion compensation value. An image stabilization device comprising a memory for storing image data on a field mountain and a controller for correcting the movement of the image data stored in the memory by the received accumulated motion vector. Convert the data into binary edge data, match the pattern of the edge data of the current field and the previous field to generate correlation data of the local motion detection area units sequentially, and analyze the generated correlation data Means for generating statistical variables and local motion vectors of the local motion detection regions sequentially, and receiving and storing the local motion vectors and statistical variables sequentially, and sequentially storing the isolation and stability of the stored local motion vectors. The weighted signals corresponding to the preset ratios Receiving means for generating a field motion vector by mixing the weights and the corresponding locally generated motion vectors, which are adaptively generated and sequentially generated, and averaging them in a field period, and receiving the field motion vector and the previous cumulative motion vector, According to the magnitude of the previous cumulative motion vector, an error motion compensation value corresponding to a preset ratio is adaptively generated, and a compensation field motion vector whose false motion compensation is suppressed by reducing the field motion vector by the error motion compensation value is determined. Means for generating the cumulative movement vector by accumulating means and receiving the compensation field motion vector and receiving the compensation field motion vector and outputting the attenuation means and accumulating the received compensation field motion vector. Image stabilization apparatus, characterized in that configured. 21. The apparatus of claim 20, wherein the means for generating a local motion vector comprises: edge detection means for receiving the image data, analyzing the received image data, and converting the field image into binary edge data; and at least two local motions. A storage means for receiving the binary edge data, sequentially extracting block data of the reference region positions of the local motion search regions, delaying one field, and outputting the edge detection means; Receive binary edge data as comparison data, receive binary edge data of the previous field outputting the memory means as reference data, pattern-match the block data of two block data, and correlate the correlation diagram data of corresponding local motion detection areas. Receiving edge pattern matching means and the correlation data, which are sequentially generated, and corresponding local motion detection operations; A locality corresponding to the position value of the maximum correlation diagram while generating the state of the correlation data which is changed by comparing the interphase diagram data received at the station with previous statistical variables detected in the corresponding local motion detection area as a statistical variable. An image stabilization device comprising local motion determining means generated by a local motion vector of a motion detection region. 22. The apparatus according to claim 21, wherein the local motion determining means generates a first comparison signal when the correlation degree data is large compared with a previous maximum correlation of the local motion detection area corresponding to the received correlation data. Receiving an address clock and an output of the comparison means, counting the address clock to generate a position value of the received correlation diagram, and corresponding to the position value of the received correlation diagram data when the first comparison signal is received. Receives local motion vector detection means generated as a local motion vector of a local motion detection region, and outputs of the received correlation data and comparison means, and generates the received correlation data when the first comparison signal is received. Means for receiving the output of the maximum correlation and generating means and the comparison means, and the previous maximum correlation when receiving the first comparison signal. Means for generating a second maximum correlation, a means for receiving the maximum correlation and a second maximum correlation, subtracting the two received signals to generate a maximum correlation, and transferring the received correlation data. Means for generating the statistical variables of the corresponding local motion detection areas, comprising means for generating the average correlation with the correlation data added and the added correlation data by the number of candidate motion vectors. . 23. The method of claim 22, wherein the means for generating the field motion vector receives the statistical variables sequentially output from the local motion determining means, and compares the statistical variables with predetermined thresholds to determine the reliability of the local motion vector. Reliability determination means for judging the occurrence of abnormal conditions and reducing abnormality and generating abnormal signals when an abnormal condition occurs, the local motion vector and the local motion vector generated in units of fields sequentially output from the local motion vector determination means, and Receiving previous field motion vectors, subtracting the motion vector information to calculate the isolation stability of the corresponding local motion vector, generating the weight signal adaptive to the set analogy according to the calculated result, and determining the reliability Means for controlling the output of the weighted signal according to the output of An image stabilization apparatus comprising: a mean weight calculation means for sequentially assigning weight signals and local motion vectors, and averaging them to generate field motion vectors. 24. The apparatus of claim 23, wherein the reliability judging means compares the average correlation with a first threshold value that is a reference value of low contrast and generates first abnormality signal when the average correlation is less than the first threshold value. and; The difference signal is generated by subtracting the maximum correlation value and the average correlation value, and the difference signal is compared with a second threshold value which is a reference value for determining a moving object. Generating second comparing means; The difference signal is generated by subtracting the maximum correlation value and the second maximum correlation value, and the difference signal is compared with a third threshold value that is a reference value for determining a repetitive image, and the difference signal is smaller than the third threshold value. And a third comparison means for generating an abnormal signal at the time of occurrence. 25. The method of claim 24, wherein the weight signal generating means receives the local motion vector and the average local motion vector, calculates an isolation degree by calculating a difference between the two motion vectors, and sets a preset value according to the calculated isolation degree. The analogy is a means for applying a weight when the isolation is large and a weight when the isolation is small to adaptively generate an isolation weight signal, and receiving the local motion vector and the previous field motion vector. Stability is calculated by calculating the difference of the motion vectors, and when the stability is large, the weight is small and the weight is large when the stability is small according to the calculated stability. Means for receiving the isolation weight signal and the stability weight signal and controlling the output of the reliability determination means. Arc received, the image stabilizing apparatus further comprise means for generating a local motion vector of the corresponding weight signal by averaging the two weights is received. 26. The apparatus of claim 25, wherein the field motion determining means comprises: means for generating weighted local motion vectors by multiplying weight signals corresponding to the sequentially received local motion vectors, and means for adding the weighted signals; And means for adding the weighted local motion vectors and dividing the added local motion vector by the added weight signal to generate a field motion vector. 27. The method of claim 26, wherein the means for generating a compensation field motion vector receives the previous cumulative motion vector, examines the magnitude of the previous cumulative motion vector, and compensates for the error motion when it has a small size at a predetermined ratio according to the magnitude. Means for adaptively generating an error motion compensation value by decreasing an error motion compensation value when the value is increased and having a large value, receiving the error motion compensation value and the field motion vector, and receiving the error motion in the field motion vector. Means for generating a first compensation field motion vector by subtracting a compensation value, receiving the error motion compensation value and the field motion vector, and adding the field motion vector and the error motion compensation value to add a second compensation field motion vector. Means for generating and comparing the error motion compensation value with the magnitude of the field motion vector, Is positive and greater than the error motion compensation value, selectively outputs the first compensation field motion vector, and if the field motion vector is positive and less than the error motion compensation value, reset the compensation field motion vector value, and reset the field motion vector. Is negative and less than the value of the error motion, the second compensation field motion vector is selectively outputted; and if the field motion vector is negative and is greater than the error motion compensation value, the compensation field motion vector is reset. Image stabilization device. 27. The method of claim 26, wherein the means for generating a compensation field motion vector receives the previous cumulative motion vector, checks the magnitude of the previous cumulative motion vector and has a small size at a preset ratio according to the magnitude. Means for adaptively generating a first error motion compensation value by increasing the motion compensation value and decreasing the error motion compensation value when it has a large value, and receiving the maximum correlation differences generated from each of the local motion detection areas. Means for generating an average maximum correlation difference of one field, and when the size of the average maximum correlation difference is examined and has a small size at a preset ratio according to the size, the error motion compensation value is increased and has a large value. Means for adaptively generating a second error motion compensation value by reducing a time error motion compensation value, the first error motion compensation value and the second error; Means for generating an error motion compensation value by averaging the current motion compensation value, receiving the averaged error motion compensation value and the field motion vector, subtracting the error motion compensation value from the field motion vector, and subtracting the first compensation field. Means for generating a motion vector, receiving the averaged error motion compensation value and the field motion vector, and adding the field motion vector and the error motion compensation value to generate a second compensation field motion vector; The motion compensation value is compared with the magnitude of the field motion vector, and if the field motion vector is positive and greater than the error motion compensation value, the first compensation field motion vector is outputted, and the field motion vector is positive and greater than the error motion compensation value. A small value resets the compensation field motion vector value, and the field motion vector is negative. And a means for selectively outputting the second compensation field motion vector when the error motion compensation value is smaller, and resetting the compensation field motion vector when the field motion vector is negative and larger than the error motion compensation value. Device. 29. The method of claim 28, wherein the means for generating the cumulative motion vector comprises: means for inputting the previous cumulative motion vector to attenuate it into a first attenuation coefficient; and inputting the previous cumulative motion vector to the first attenuation coefficient; Means for attenuating with a second attenuation coefficient having a smaller value, receiving the field motion vector and correlation data, and analyzing the field motion vector and correlation data to consider panning when a motion vector in the same direction is generated over a predetermined frame. Panning identifying means for generating a panning signal; Receives outputs of the first attenuating means and the second attenuating means and receives an output of the panning identification means as a selection signal, selects an output of the second attenuating means upon receiving the panning signal, or else the first attenuation means And a means for selecting an output of the means, a means for receiving the compensation field motion vector and the output of the selection means, generating the cumulative motion vector by adding the two signals, and outputting the accumulated motion vector to the controller. Device. 29. The method of claim 28, wherein the means for generating the cumulative motion vector comprises: means for inputting the previous cumulative motion vector to attenuate it into a first attenuation coefficient; and inputting the previous cumulative motion vector to the first attenuation coefficient; Means for attenuating a second attenuation coefficient having a smaller value, receiving the field motion vector and the correlation diagram data, and analyzing the field motion vector and the correlation diagram data for panning when a motion vector in the same direction is generated over a predetermined frame. A panning identification means for generating a panning signal, and receiving outputs of the first attenuation means and the second attenuation means and receiving an output of the panning identification means as a selection signal, and receiving the second attenuation means when receiving the panning signal. Means for selecting the output of the first attenuation means, and receiving the compensation field motion vector and the output of the selection means. And a means for generating a cumulative motion vector by adding the two signals, and between the adding means and the control unit. When the cumulative motion vector exceeds a maximum correction region of the motion vector, the cumulative motion vector is set to a constant value. Image stabilization apparatus comprising a limiter to maintain. An adaptive motion vector determination method of a digital image stabilization system comprising a memory for storing image data in field units and a controller for correcting the movement of the image data stored in the memory based on the received accumulated motion vector. A process of converting the image data in field units into binary edge data, extracting the reference block data of the local motion detection area by receiving the binary edge data, and sequentially storing the data in the corresponding reference data storage areas. And sequentially receiving the binary edge data of the converted current field as comparison block data and pattern matching the reference block data of the corresponding local motion detection area to sequentially generate correlation diagram data of the local motion detection area units. And a corresponding local part by analyzing the sequentially received correlation diagram data. Generating statistical variables and local motion vectors of direct detection areas, and resetting upon receiving a search termination signal of the local motion areas; sequentially storing the received local motion vectors and statistical variables; Upon receiving the field search termination signal, the stored local motion vectors are sequentially examined to generate weight signals of isolation and stability, and the statistical variables are sequentially analyzed to analyze the state of the image signal to control the output of the weight signals. And multiplying the weighted signals corresponding to the sequentially received local motion vectors, and averaging the signals at field periods to generate a field motion vector, and attenuating a previous cumulative motion vector at a predetermined ratio. Add the previous cumulative motion vector and the field motion vector W image stabilizer characterized in that constituted by any process for the generation of a cumulative motion vector. An adaptive motion vector determination method of a digital image stabilization system comprising a memory for storing image data in field units and a controller for correcting a motion of the image data stored in the memory based on a received compensation accumulation motion vector. Converting binary edge data from received field data, extracting the reference block data of the local motion detection area by receiving the binary edge data, and sequentially storing the data in the corresponding reference data storage areas. Process and receiving binary edge data of the current field to be converted as comparison block data, and matching the reference block data of the corresponding local motion detection area to sequentially generate correlation diagram data of the local motion detection area units. Process and the correlation diagram data received sequentially Generating statistical variables and local motion vectors of the boolean motion detection regions, and storing the received local motion vectors and statistical variables in a sequential order; Upon receiving the field search termination signal, the stored local motion vectors are sequentially examined to generate weight signals of isolation and stability, and the statistical variables are sequentially analyzed to analyze the state of the image signal to output the weight signal. A process of controlling, multiplying the weighted signals corresponding to the sequentially received local motion vectors, averaging these signals by field periods, generating a field motion vector, and attenuating a previous cumulative motion vector at a predetermined ratio The reduced previous cumulative motion vector and the field motion vector Calculating the cumulative motion vector, analyzing the magnitude of the cumulative motion vector, generating an error motion compensation value according to a preset ratio, and reducing the cumulative motion vector by the error motion compensation value to obtain an error motion vector. An image stabilization apparatus comprising a process of generating a compensation cumulative motion vector to suppress. An adaptive motion vector determination method of a digital image stabilization system comprising a memory for storing image data in field units and a controller for correcting a motion of image data stored in the memory based on a received cumulative motion vector. Converting image data in field units into binary edge data, extracting the reference block data of the local motion detection region by receiving the binary edge data, and sequentially storing the data in the corresponding reference data storage regions And receiving, by the converted field, binary edge data as a comparison block data, pattern matching the corresponding block data of the corresponding local motion detection region, and generating locality correlation data sequentially by the local motion detection region units. The corresponding local movement is analyzed by analyzing the received correlation diagram data sequentially. Generating statistical variables and local motion vectors of the detection areas, resetting upon receiving the search termination signal of the local motion area, sequentially storing the received local motion vectors and statistical variables, and Upon receiving the search termination signal, the stored local motion vectors are sequentially examined to generate weight signals of isolation and stability, and the statistical variables are sequentially analyzed to analyze the state of the image signal to control the output of the weight signals. And multiplying the weighted signals corresponding to the sequentially received local motion vectors, averaging these periods to generate a field motion vector, and analyzing a magnitude of a previous cumulative motion vector according to a preset ratio. Generate an error motion compensation value and generate the error motion compensation value Generating a compensation field motion vector that suppresses the error motion vector by reducing the field motion vector, attenuating a previous cumulative motion vector at a preset ratio, and reducing the attenuated previous motion vector and the compensation field motion vector. And adding the cumulative motion vector to the image stabilization method.