JPH05336374A - Image information compressing device - Google Patents

Abstract

PURPOSE:To decrease data quantity while maintaining picture quality. CONSTITUTION:This device is composed of memory(MEM) 105 which holds inputted image data(DT) transiently, means 110, 118 which set the remarked area of an image on the memory 105 and other areas, a separation means 111 which distinguishes the area from the HEM classified by every set area and reads it out in block(BL) unit of prescribed pixel size, a quantization table 123 provided with plural pairs of tables(TBL) for quantization with different compressibility, a means 119 which discriminates the detail of the image in a read out BL, means 122, 124 which select the TBL defined corresponding to a discrimination result and the prescribed area of the read out BL from the quantization table, a means 113 which applies orthogonal transformation(DCT) to the image data read out by a reading means, a means 114 which performs the quantization on DCTed data by using a selected TBL, a means 115 which applies encode processing to quantized data, and a means 116 which multiplexes and outputs encoded data with position information and information for selection.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image information compression apparatus capable of storing a facial photograph at a high compression rate while maintaining the image quality required for printing.

[0002]

2. Description of the Related Art Image information has a large amount of information as compared with character and numerical information, so that there are many restrictions on the storage capacity, processing time, etc., and therefore its use is limited. However, in recent years, such restrictions are being removed along with the progress and development of image information devices.

By the way, one of the fields in which an image plays an important role is an ID (identification) field. Particularly in the field of personal identification, the importance of facial photographs has long been recognized, and it has been used for ID purposes such as silver lead photographs and instant photographs.

However, in the information of such printing paper,
The information exists only in the print itself, and it cannot be converted into an electronic file, processed, or copied, and its application is extremely limited.

In order to eliminate such a problem, it is known to convert face photograph information into an electric signal and handle it. For example, FIG. 13 is cited from Japanese Patent Laid-Open No. 63-316275.

FIG. 1 shows an ID card producing device,
In this device, the subject 1 is imaged by the color TV camera 2,
The image signal thus obtained is converted into a digital multi-gradation signal by the A / D conversion circuit 3 and temporarily recorded in the frame memory 4.

Reference numeral 5 denotes an image signal processing circuit, which performs calculations for removing color turbidity due to an incorrect spectral absorption element of a color material applied to a full-color video printer. Reference numeral 6 is a controller for controlling the operation of the entire apparatus, 7 is an input keyboard, and 8 is a full-color video printer unit.

The keyboard 7 is operated, whereby the controller 6 generates a graphic image or a character image, which is combined with the human image of the subject 1 stored in the frame memory 4 and edited to form an ID card screen. It is converted into a TV image signal and then displayed on a CRT, thereby converted into an optical image, which is imaged on a color photographic paper by an optical system and exposed. Then, the exposed printing paper is developed and printed out.

According to this system, a subject image can be processed, edited and printed together with character information and the like, and necessary information including the subject image can be printed on one card to form an ID card.

By the way, in another example of using a face photograph,
There is a database for managing the information of members who belong to organizations such as companies and schools. As an example of such a database system, the one shown in FIG. 14 is known.

According to this apparatus, an image signal obtained from a video image pickup apparatus 10 such as a still video recording player, a video camera, a still video camera or a video presentation camera is taken in via an image processor 11 and displayed on a monitor 12. Further, it can be recorded in the data file 13 and can be searched and displayed later, or can be printed out by the video printer 14.

In this apparatus, an image signal is processed by the image processor 11 and then recorded and stored in the data file 13, but since it is stored as a video image, its data capacity becomes a problem. That is, in a video image, the amount of information of a facial photograph image is required to be several 100 Kbytes per person, and when a general external storage device such as a magneto-optical recording disk or a hard disk is used as the data file 13, such an external storage is required. Since the memory capacity of the device is at most several 100 Mbytes to several Gbytes, it will be filled with about 1000 people. For this reason, a high-efficiency image encoding technique is introduced to compress the data amount per image, enabling a larger number of images to be stored.

In the past, DPC was used as an encoding method.
M (predictive coding) was used, but the compression rate was poor at about 1/2, so coding that enables even higher compression rates in response to increasing information volume and increasing transmission speed Technology is needed.

Vector quantization and DCT (discrete cosine transform) are mainly used as high-efficiency coding that meets such requirements.
A coding method using is being studied, and image data compression technology based on these highly efficient coding methods has been introduced in image database file systems in recent years.

[0015]

In the conventional system in which the data of the face photograph image is stored in the database and used as described above, the stored face photograph image is searched and referred to, or printed for the ID card. It can be used in different ways. Therefore, in consideration of these matters, it is necessary to file the image at an image quality level that matches the resolution of the printer used in the system. In a general image filing system, as long as a CRT image is viewed, it is usual to perform filing processing by increasing the image compression rate so that deterioration of image quality is not a concern.

However, in image compression aiming at the image quality of a CRT image, even if there is no problem in image quality as far as the CRT image is seen, it cannot be used in a print image which is required to have much higher image quality. It could lead to such a situation.

That is, since a general CRT display device has about 525 scanning lines and has a maximum horizontal resolution of only about 300 lines, it is meaningless to use very fine image data for the image itself which can be displayed on the CRT. .. Therefore, in the case of an image used for CRT display, there is no problem even if the image is compressed at a high compression rate, but in the case of a printed image, the fineness of the texture is directly reflected on the image quality.

When a processed image with a high compression ratio is printed, block distortion becomes noticeable in a reproduced print image in a compressed image adopting the DCT method as the compression method, and the vector quantization method is used for compression. In that case, the resolution is greatly deteriorated, resulting in a blurred image. This tendency becomes remarkable as the compression rate is increased.

For this reason, the compression rate is often kept low, and in some cases, the image is recorded and stored without being compressed. Therefore, the image compression technology has not been fully utilized in the system, such as limiting the number of face images to be recorded and stored, and preparing a large memory capacity to support the majority.

Further, there is software (such as Storm Technolegy's product name "Picture Press") that divides an area within one screen and compresses each area at a different image quality level, but it is suitable for rough processing. However, the detailed processing could not be performed.

Therefore, the development of a system capable of reducing the amount of image data per person while maintaining the image quality level usable for printing and enabling the recording and storage of a larger number of facial photograph images has been awaited.

An object of the present invention is to provide an image compression apparatus and a reproduction apparatus thereof capable of recording and storing a large number of images by reducing the amount of image data per sheet while maintaining the image quality level usable for printing. To provide.

[0023]

In order to achieve the above object, the present invention is configured as follows. That is, first, an image data input means for inputting image data of a subject image, an image memory for temporarily holding the input image data, an attention area of the image on the image memory, and other areas are set. Setting means and separating means for dividing the set attention area and the other areas and reading out from the image memory for each area in block units of a predetermined pixel size;
A quantization table having a plurality of sets of tables for quantization with different compression rates, a determination means for determining the fineness of the image of the read block, and a determination result of the determination means. Selection means for selecting from the quantization table a compression rate table determined corresponding to the area to which the read block belongs, orthogonal transformation means for orthogonally transforming the image data read by the reading means, and the orthogonal transformation means. Quantizing means for quantizing the data obtained by conversion using the table selected by the selecting means,
Coding means for coding the quantized data, and multiplexing for multiplexing and outputting the coded data together with position information at the time of reading by the reading means and information for selecting the used table. The selection criterion of the table is defined such that the image of the region of interest is compressed at a lower compression rate as the image is finer and the images of the other regions are compressed at a higher compression rate. ..

Secondly, the image data of the subject image is held in the image memory for temporarily holding the image data, and the image data in the image memory is divided into a set attention area and other areas, and each of the areas is predetermined. This is read out in block units of pixel size, subjected to orthogonal transformation, and the data obtained by this transformation is quantized with a predetermined quantization table according to the position on the image memory and the state of the image of the block. Temporarily holds image data as a reproduction device of compressed and recorded image information in which the quantized data is encoded and recorded together with position information on the image memory of the block and information for selecting the used quantization table. Image memory, quantization table with multiple sets of quantization tables with different compression ratios, and input compressed recorded image Separation means that separates and extracts the encoded data from the information, the position information at the time of reading by the reading means, and the information for selecting the used table, and the encoding means that is separated by the separating means. Decoding means for decoding the generated data, and selection means for selecting a quantization table determined by the information from the quantization table by using information for selecting the table separated by the separating means. And an inverse quantizing means for inversely quantizing the data inversely encoded by the inverse encoding means using the quantization table, and inversely orthogonally transforming the inversely quantized data into image data. Inverse orthogonal transforming means, synthesizing means for writing the image data on the image memory corresponding to the position information separated by the separating means, and on the image memory Was more configuration and output means for outputting the image data read Te as restored image signal.

[0025]

In such a structure, in the case of the first structure,
The image data of the subject image is held in the image memory that temporarily holds the image data, and the image data on the image memory is divided into the set attention area and the other area and read in block units each having a predetermined pixel size. Is orthogonally transformed, and the data obtained by this transformation is quantized by a predetermined quantization table according to the position on the image memory and the state (fineness) of the image of the block, and the quantized data is quantized. Is encoded and superimposed with the position information on the image memory of the block and the information for selecting the used quantization table to create compressed recording image information.

In the quantization, the selection criterion of the table is defined so that the image in the region of interest is compressed at a lower compression rate as the image is finer and the images in the other regions are compressed at a higher compression rate. I chose

Therefore, since the image of the region of interest has a low compression rate, there is no fear of deterioration even if the image is expanded, and since the compression rate of other regions is high, the part may deteriorate when the image is expanded. Since the image quality in the attention area is ensured, by setting the attention area in the face portion like the ID image, the image quality in the necessary portion is maintained and the high data compression in other portions is performed. Due to the effect, the amount of image data can be reduced. Therefore, according to the present invention, the amount of image data per sheet can be reduced while maintaining the image quality level usable for printing, and a large number of images can be recorded and stored.

Further, according to the second configuration, the quantization table is provided with a plurality of sets of tables for quantization in which the magnitudes of the compression rates are different, and the separating means inputs the compressed recorded image. From the information, the encoded data, the position information at the time of reading by the reading means, and the information for selecting the used table are separated and taken out, and the encoded data among the separated data is extracted. Is inversely encoded by the inverse encoding means. The selecting means uses the information for selecting the table among the separated data, selects the quantization table determined by the information from the quantization table, and the dequantizing means selects the selected quantum. The inversely encoded data is inversely quantized by using the conversion table. Then, the inverse orthogonal transform means inversely orthogonally transforms the dequantized data into image data, and the synthesizing means associates the image data with the position information separated by the separating means and stores it in the image memory. Write. Then, the output means reads the image data on the image memory and outputs it as a restored image signal. As a result, the image data compressed by the device having the first configuration can be restored.

[0029]

Embodiments of the present invention will be described below with reference to the drawings. The present invention divides an image into areas of a face portion and the other image portion so that the sharpness is as inferior as possible in the face portion, and high compression is possible in the non-face portion even if the sharpness is slightly inferior. In addition, a first compression rate weight table is provided separately for that area. The resolution of each block after the DCT calculation is calculated from the four DCT coefficients, and the classification is performed. A second compression rate weight table is provided for each block according to the class value. Also, a plurality of compression quantization tables with different compression rates are provided so that one of them can be selected when performing quantization. This selection is made based on the product value of the weight values of the first and second tables. As a result, a different quantization table is selected for each block, and compression processing can be performed at a compression rate according to the image state of the area and block. When the compressed data is encoded and the encoded data is stored, the product value and the area data described above are recorded and stored together. The decompression process is the reverse process and is performed by the same means. In addition, after restoration, the configuration is such that images can be combined.

A specific embodiment for realizing such processing will be described below. (First Embodiment) FIG. 1 is a block diagram of an embodiment for realizing the present invention. In this figure, 101 is a subject such as a person, 102 is a background portion behind the subject 101,
3 is a color TV for shooting the upper half of the subject 101
A camera (color video camera) 104 converts an analog signal (R (red), G (green), B (blue) color video signal of image) which is an output signal from the color TV camera 103 into a digital signal. It is an A / D converter that outputs the output.

Further, 105 is a frame memory, which is an A / D
The signal data output from the converter 104 is temporarily stored as a frame image, and 106 is a D / A converter for converting the digital signal output from the frame memory 105 into an analog video signal.

Reference numeral 107 is an output signal conversion unit for converting an output signal from the D / A converter 106 or an output signal from the color TV camera 103 into, for example, an NTSC system video signal, and 108 is an output signal conversion. It is a TV monitor for displaying a video signal converted and output by the unit 107 as an image.

Further, 109 is a controller that controls the entire system, and 110 is a keyboard, which allows the operator to input commands, input character information, and perform editing operations on this system. It is for doing. The input from the keyboard 110 is given to the controller 109. The keyboard 110 is also used for inputting coordinate values and various instructions to be described later.

Reference numeral 111 denotes a face / non-face area separation / synthesis unit, which performs a process of cutting out image data into a face portion and a non-face portion in the image data in the frame memory 105 or performing both synthesis. To do.

An image processing unit 112 cuts out or combines a block image of 8 pixels × 8 pixels for image compression, and various images such as color conversion called RGB video signal / YUV video signal conversion. It is something that can be processed.

Reference numeral 113 denotes a DC capable of performing discrete cosine transform (DCT) which is one of orthogonal transforms and its inverse transform.
A T processing unit and a quantization unit 114. The quantizer 114 uses the quantization table 123 to perform quantization or inverse quantization with a different step size for each DCT coefficient.

Reference numeral 115 is an encoding unit for performing variable length encoding processing such as run length processing and Huffman encoding performed by rearranging quantized data, or inverse encoding processing, and 116 is an encoding unit. A multiplexing unit 117 that multiplexes the data encoded in 115 and compressed area data described below, and decodes the multiplexed data, is denoted by 117. For example, a magneto-optical device for recording and storing the compressed data and area data for input / output. It is an external storage device such as a disk device or a hard disk device.

Reference numeral 118 denotes a face / non-face area setting unit that allows the operator to arbitrarily set the area division of the face area and the non-face area in the image in the frame memory 105. The part / non-face part area setting unit 118 further inputs / outputs area data to / from the external storage device 117 so that the operator can set the predetermined value in consideration of the operability of the operator. It is configured to allow transfer.

Reference numeral 119 denotes a resolution discriminating unit which evaluates the degree of fineness for each block based on the added value of four coefficients among the DCT coefficients calculated by the DCT processing unit 113 and classifies the degree according to the degree. Yes, 120 is a compression rate weight table for each block used for digitizing the degree of compression rate as a weight from the values classified into the classes,
Reference numeral 121 denotes an area-specific compression rate weight table that receives information on whether the processing area is a face area or a non-face area from the face / non-face area separation / combination unit 111 and digitizes the value from the value as the compression rate degree. ..

Reference numeral 122 denotes a compression rate setting and area reading section, which calculates the product of the compression rate weight table value for each area and the compression rate weight table value for each block at the block position for compression processing, and the compression rate is based on the product. Is set, and the data of the position information is input and output. Reference numeral 123 is a quantization table composed of a plurality of tables with different compression rates. Reference numeral 124 selects one specific table corresponding to the output value from the quantization table 123 corresponding to the output value of the compression rate setting and area reading unit 122,
It is a table selection unit provided to the quantization unit 114.

Next, the operation of the system having the above configuration will be described. The upper body including the face of the person of the subject 101 is photographed by the color TV camera 103. At that time, in order to make it easier to see the person, a monotone background portion 10
Place 2 and shoot.

The camera 103 is previously adjusted for white balance, lens aperture adjustment, and focus adjustment. The RGB analog video signals output from the camera 103 are converted into digital signals by the A / D converter 104 and sequentially recorded in the frame memory 105.

The digital data stored in the frame memory 105 in this manner is sequentially read out in synchronization so that it can be output as a video signal, and the D / A converter 106 and the output signal conversion unit 107 perform analog RGB video signals. Convert to. This signal is sent to the TV monitor 10
By inputting in 8, the upper half image of the person who is the subject can be monitored. The operator adjusts the illumination and the camera so that the face image can be properly taken while looking at the image on the TV monitor 108.

A face image used as an ID image is photographed as follows. First, the image data being taken is captured in the frame memory 105. Therefore, the frame memory 105
In order to freeze (freeze) the face image above, the controller 109 sends a freeze control signal 105
To the frame memory of.

In order to perform this operation, the operator inputs a key for the freeze from the keyboard 110 at the moment when the operator wants the freeze. As a result, the controller 109
Receives a signal from the keyboard 110, stops the sequential write data address counter of the frame memory 105, and stops writing. But the controller 1
With regard to the read data address control, by continuing the operation as it is, the image data in the frame memory 105 is controlled to be read regardless of the presence or absence of the freeze operation.

The image data read from the frame memory 105 can be output as a video signal by transferring to the D / A converter 106. Therefore, by providing the video signal output in this manner to the TV monitor 108 and displaying it, the moving image or still image of the camera image can be observed. When obtaining the ID image, the operator performs an image area dividing operation in this state.

In the image area dividing operation, the area side including the face is set to the face portion (B portion), and the remaining image area is set to the non-face portion (A portion).
Can be performed by specifying the area by the face / non-face area setting unit 118.

For example, taking the case of area division as shown in FIG. 2, the image coordinates (x, y) are first determined for convenience. The coordinate positions of the outermost corners (four corners) of the image are (x _A1 , y _A1 ), (x _A2 , y _A1 ), (x _A1 ,
y _A2 ), (x _A2 , y _A2 ), and the external angle coordinates of the face area B part are (x _B1 , y _B1 ), (x _B2 , y _B1 ), (x _B1 , y _B2 ),
(X _B2 , y _B2 ). These area separation / combination are performed by the face / non-face area separation / combination section 111, and the areas and coordinates thereof can be set by the face / non-face area setting section 118 for the image on the frame memory 105. I am trying.

In the state before setting these coordinates, first, the controller 109 sets the face portion and the non-face portion so that the image of the frame C having a single color standard size as shown in FIG. 2 can be superimposed on the image on the frame memory 105. The area separating / combining unit 111 is controlled, whereby the face / non-face area separating / composing unit 11 is controlled.
1 is a frame memory 10 in which the set standard size frame C is set.
5 produces an image on top.

That is, in order to display the frame C at a fixed position in the standard size, the corner position coordinate data of the standard frame C is set in the face / non-face area setting unit 118, and the set value is also set. A frame image is created under the control of the controller 109 and stored in the frame memory 105. The person image and the frame image may be stored in the same frame, but by storing them in different frames and combining them when reading, an image in which the frame image and the person image are superimposed may be obtained. good.

In this way, the image on which the frame C is superimposed is T
The image is displayed on the V monitor 108, and while monitoring this, the operator confirms whether the frame C is properly positioned with respect to the face portion of the person. If it is not correct, the operator uses the keyboard 1
By operating 10 and inputting new coordinate values (x _B1 , y _B1 ) and (x _B2 , y _B2 ) indicating the diagonal position of the region to be cut out, the controller 109 outputs this input data to the face / non-face part. It is given to the face area setting unit 118 and set.

As a result, the face / non-face area setting unit 11
In 8 is written the coordinate data of a new frame C surrounding a rectangular area whose diagonal position is the coordinate position determined by this coordinate data, and the image of the new frame C is created based on the value of the new coordinate data. Since it is generated, the subject image can be monitored on the TV monitor 108 with the face portion surrounded by the new frame C.

By repeating this, the operator can adjust the cutout area while looking at the relationship with the frame C so that the image of the face portion (B portion) of the person is properly arranged in the frame C.

When the proper cutout area is set in this way, thereafter, cutout processing is started in order to record the area A and the area B with different image quality, and a series of processing such as compression is automatically performed. enter. The cutout is executed by the controller 109. The controller 109 cuts out the outside of the B area and then cuts out the inside area.

Regarding the clipping of the external area, the image data corresponding to the area can be read out from the frame memory 105, and the image data corresponding to the area can be read out from the frame memory 105 for the internal area. 10
5, the addresses of the frame memory 105 corresponding to the respective areas are sequentially generated, and the image data is read from the frame memory 105.

That is, the cutout of the area A is carried out at coordinate positions (x _A1 , y _A1 ), (x _A2 , y _A1 ), (x _A1 , y _A2 ),
Regarding the range surrounded by (x _A2 , y _A2 ), it is the B part area (x _B1 , y _B1 ), (x _B2 , y _B1 ), (x _B1 , y _B2 ),
Except for the range surrounded by (x _B2 , y _B2 ), the address control is performed to read out the data from the frame memory 105, the data is transferred to the image processing unit 112, and the B area is cut out at the coordinate position (x _B1 , y _B1 ),
In order to read the range surrounded by (x _B2 , y _B1 ), (x _B1 , y _B2 ), and (x _B2 , y _B2 ), address control is performed to read data from the frame memory 105 and transfer it to the image processing unit 112. By doing. This is done separately for each region.

Area A and area B cut out in this way
Each image in the sub-region is compressed by performing compression processing. The compression is based on the DCT method.
The image data of the area A or area B that has been cut out and transferred is further cut out by the image processing unit 112 into a block image of 8 dots × 8 dots, and the following processing is performed for each block.

First, an image for each RGB component of the block is converted into a YUV image including a luminance component and other two colors (chroma blue component, chroma red component, color difference component signal) data. This is performed by the image processing unit 112. After that, the respective image data are processed by the DCT processing unit 11
In step 3, DCT (discrete cosine transform) is performed.

By this DCT, the spatial frequency components in the block image are divided for each frequency component and obtained as DCT coefficients. Then, with respect to the DCT coefficient for each frequency component, the quantization unit 11 is based on one table (for luminance and for the other two colors) of the plurality of quantization tables 123.
4 is quantized with a different step size for each DCT coefficient for each frequency component. Such quantization is performed for each of the luminance component and the other two color components. Then, the quantized data is subjected to run length processing in the coding unit 115 and further subjected to variable length coding processing to obtain one block of compressed image data. This data and compressed area data to be described later are multiplexed signals by the multiplexing unit 116, transferred to the external storage device 117, and recorded as an image data file.

When decompressing (decompressing an image) the compressed image data, the reverse process described above is performed.
First, the image data file is read from the external storage device 117, data is transferred to the multiplexing unit 116, and the data of the multiplexed image data file is decoded into compressed data and compressed area data.

The compression unit 115 returns the compressed data to the original encoding process and the data array, and the quantization unit 114 further performs the inverse quantization process. The quantization table used in the inverse quantization process can be selectively controlled from the compression area data described later so that the same table as the table used in the compression is selected.

By the way, the luminance data (Y signal data) and the other two color data (data for the color difference signals) are inversely quantized, and the DCT processing unit 113 performs inverse DCT processing. This gives the original 8 dots x 8
It is converted back to YUV image data of dots.

The 8 dot × 8 dot YUV image data obtained by being inversely transformed by the DCT processing unit 113 is transferred to the image processing unit 112, and this image processing unit 1
In 12, the YUV image data is converted into RGB image data to obtain block image data. In this way, the compressed image is restored.

The restored image data is combined and written in the frame memory 105 by the following processing. That is, based on the compressed area data that is the area setting information recorded in the compressed image file, the controller 109 sets the information on the area A and the area B as coordinate information in the face / non-face area setting unit 118. Then, the face / non-face area separation / combination unit 111 takes in the data of the restored block image from the image processing unit 112, and based on the information set in the face / non-face area setting unit 118. The image data is written pixel by pixel on the frame memory 105 so that the data of the area A is located at the corresponding position of the area A and the data of the area B is located at the corresponding position of the area B. Images of face and non-face are combined.

Then, the image data on the frame memory 105 is sequentially read out and converted into an analog signal by the D / A converter 106, and the output signal converter 107 outputs NT.
After being converted into an SC system video signal, it is given to the monitor 108 and displayed as an image. It is restored in this way. Next, selection of the quantization table 123 will be described.

The map data in which the values related to the compression rate are assigned to each of the above 8 × 8 pixel blocks is set so that the processing can be performed for each block while changing the compression rate. This map data will be referred to as compressed area data here.

The map data is created by the product value of two tables. One table is the area-based compression rate weight table 121, and the other table is the block-based compression rate weight table 120. Area-based compression ratio weight table 12
Regarding No. 1, the face part can be compressed while maintaining the sharpness as much as possible without increasing the compression rate, that is, the non-face part, and the background part that does not require a high degree of detail. Since it is the part excluding the face of the person, the compression ratio is increased, that is, the degree of fineness is reduced, but large compression can be performed to reduce the data amount. Use rate weight table.

The larger the weighting coefficient, the higher the compression. The area corresponds to FIG. 2, and the part A is w _A ,
The portion B is set to w _B , for example, w _A = 2 and w _B = 1. With respect to the compression rate weighting table 120 for each block, the DCT coefficient obtained by the DCT processing in the DCT processing unit 113 is extracted, the degree of fineness of the image is determined based on the added value, and the image is classified into classes. If the degree of fineness of the image is high, the compression rate is set low to maintain the degree of fineness, and if the degree of image fineness is low, the compression rate weight value is blocked so that the compression rate can be set high. A block-by-block compression rate weight table is obtained for each.

DC subjected to DCT operation in the DCT processing unit 113
The T coefficient is shown in FIG. Of these coefficients, F in the horizontal direction
₀₅ and F ₀₆ and the absolute value of the value obtained by adding it to 4 values of F ₅₀ and F ₆₀ in the vertical direction

[0070]

[Equation 1] Or add the absolute value of each coefficient value

[0071]

[Equation 2] Ask for.

Here, for example, the latter method is used, and the resolution discrimination unit 119 calculates this. The added value is divided into integers divided by 8, and the value determined as the degree of image fineness in this block is output and received by the block-by-block compression rate weight table 120. The block compression rate weight table 120 further divides this value into, for example,
A weight table for the compression rate for each block is provided in the way of division as shown in FIG.

This value is used as the compression ratio weight table for each block 1
20 and outputs the weight value w to the compression rate setting and area reading unit 122. The compression ratio setting / area reading unit 122 uses the compression ratio weight table 12 for each block.
A product W of 0 and each weight value of the compression ratio weight table 121 for each area is calculated. This value of W will be called compressed area data.

By transferring the W value, which is the value of the compressed area data, to the table selection unit 124, control is performed so that one of the plurality of quantization tables 123 can be selected in the table shown in the table of FIG. .. As a result, the quantization table used by the quantization unit 114 is selected. These are processed block by block.

The table number in FIG. 6 is set so that the higher the number, the higher the compression processing can be performed. In this way, it is possible to perform compression processing by changing the compression rate for each area and restore it.

Next, let's consider how these compression processes can be processed differently in the partial parts of the image. For example, consider blocks BLK1 to BLK4 of 8 × 8 pixels. It is assumed that, of these, the one in the B area, which is the face area, and the one in the A area, which is the background area, are picked up as follows.

In this system, the selection of the quantization table is
The weighting index w is determined depending on whether the image of the block has a high spatial frequency or a low spatial frequency, and which quantization table is used depending on how many values are corrected by the correction constant. decide.

Whether the block image has a high spatial frequency or a low spatial frequency is determined by the resolution discriminating unit 119 which discriminates the fineness of the image as a specific numerical value according to the fineness of the image. A table selection value uniquely obtained by obtaining a weight index w from the block-by-block compression ratio weight table 120 based on the determination output and multiplying this by the correction constant output from the region-by-region compression ratio weight table 121. Determines the quantization table to be used with.

The compression ratio weight table 121 for each area is
Using the address information output from the non-face area separation / synthesis unit 111, whether the processing is currently being performed in the A area based on the situation,
Since it can be known whether the processing is performed in the B area, the weight value (area-specific compression ratio weight table value) is output accordingly.

The compression ratio weight table values for each area are w _A and w
_If there is _B and is in the B area, use w _B. Also, w _A is used for those in the area A. Then, as for the area B, as shown in FIG.
The compression rate weight table value w _B for each area is set to “1”, and the compression rate weight table value w _A for each area is set to “2” for the area _A.
And

Now, the block BLK1 is completely in the face part in the B part area, the BLK2 is in the B part area, but it is across the face part and the background, and the BLK3 is in the A part area. And BLK4 is in the background in the area A.

Since BLK1 is a face portion in the B area, the compression ratio weight table value for each area is w _B (=
“1”), and the face image has a large change in luminance and color and a monotonous part. Here, assuming that the change is large, if the output value of the resolution determination unit 119 is “12”, The output of the block-by-block compression ratio weight table 120 shown in FIG. 5 is a weight w = 1 for “12”, and the product value W thereof is W = w _B × w = 1 × 1 = 1,
From FIG. 6, the quantization table number is “1”.

In BLK2, the region is the B region, the compression ratio weight table value for each region is w _B (= “1”), and the background is monotone, so the output value of the resolution determination unit 119 is “1”. Then, the output value of the block-by-block compression rate weight table 120 is the weight w = 6, and the product value W = w
_The quantization table number is “3” when _B × w = 1 × 6 = 6.

Since BLK3 is the area A, the compression ratio weight table value for each area is w _A (= “2”), the clothes are slightly changed, and the output value of the resolution determination unit 119 is “8”. Then, the compression rate weight table for each block 120
The output value of is a weight w = 2, and the product value W = w _A × w = 2 ×
When 2 = 4, the quantization table number is "2".

Since BLK4 is the A area, the compression ratio weight table value for each area is w _A (= “2”), and the output value of the resolution determination section 119 is “1” because the background is monotone. Then, the output value of the compression rate weighting table for each block 120 is the weight w = 6, and the product value W = w _A ×
When w = 2 × 6 = 12, the quantization table number is “4”.

When the quantization table number is determined in this way, the selection information of this determined number is output from the compression rate setting and area reading unit 122 to the table selection unit 124, and the quantization table number is selected from the plurality of quantization table groups. The quantization table having the determined number is selected by the table selection unit 124 from the quantization table 123. And
The quantization table of that number is sent to the quantization unit 114,
The quantizer 114 quantizes using the given quantization table. The product value W is also sent to the multiplexing unit 116 as information necessary for decompression, is multiplexed with the compressed image data, and is stored in the external memory 117.

In this way, the processing block is the face area (B
The compression ratio is determined while determining the difference between the partial area) and the non-face area (A area) and the difference in the fineness of the image in the block, and the quantization table is selected for each block. As a result of the compression, in the part B area, there is little or little compression in the part where there is a small change, and in the part B area where there is little change, the compression ratio is relatively high, and in the part A area. The compression ratio is changed according to the area and the image fineness, such that the part with a small change is compressed with a relatively small compression rate, and the part with a small change is compressed with a large compression rate. However, the compression rate at the time of quantization will be controlled.

As for the compressed area data value W for each block, when the compressed image data is read from the external storage device 117, various information necessary for restoration is separated and reproduced for each block and output from the multiplexing unit 116, and the product value W Is obtained, and this is input to the table selection unit 124. Therefore, based on this, the compression rate setting and area reading unit 122 outputs information for the quantization table selection command according to the product value W and controls the table selection unit 124. A table corresponding to the product value W is selected from the quantization table 123 and given to the quantization unit 114.

As described above, the coordinate data (region data) obtained by dividing the image on the frame memory 105 into the face portion and the non-face portion is stored together with the compressed image data and stored in the external storage device 117. Since the data is separated and extracted at the time of reading and is made available for use, the area data is compressed and expanded and recombined to the face / non-face area setting unit 118 from the external storage device 117. The image data is transferred and the decompressed block image data output from the image processing unit 112 is fetched by the face / faced area separation / synthesis unit 111 based on the data, and the frame memory 1
By writing the decompressed block image data in the frame memory 105 while controlling the write address of 05, the decompressed block image data is correctly arranged on the frame memory 105 at the coordinate position where the block should be placed. As a result, the face image and the non-face image are restored to one original image.

Although the above embodiment describes processing using four DCT coefficients in order to determine the fineness of an image, each component in a horizontal and vertical image as shown in FIG. 7 is crossed. It is also possible to process by replacing the combined coefficients of F ₅₅ , F ₅₆ , F ₆₅ , and F ₆₆ . Further, as shown in FIG. 8, if the processing is performed with a frequency range, the size of the image information amount can be determined.

[0091]

[Equation 3] It is also possible to determine the degree of fineness of the image in which high frequency components are emphasized by weighting.

At this time, at the image processing unit 112, Y
Converted from UV data to RGB data, but YUV
Face / non-face area separation / synthesis unit 1 for Y data only
It is also possible to write in the frame memory 105 under the address control of 11, and in that case, the controller 10 can write only the face portion at an arbitrary position in the frame memory 105.
At 9 the coordinate data is rewritten in the face / non-face area setting unit 118, the face / non-face area separation / synthesis unit 111 takes out the Y data in the image processing unit 112, and the frame memory 1
Write in 05. By doing this for the data of a plurality of people, a monochrome multi-image can be displayed on the TV monitor.

In addition, the output of the output signal converter 107 is shown in FIG.
It is also possible to connect to a video printer as described above and print out as an ID card image. In this case, data such as a logo image such as a CG (computer graphic) image and a name image are stored in advance in the external storage device 117, and are connected as shown in FIG.

At the time of printing, first, the data of the logo image or the name image is transferred from the external storage device 117 to the image processing unit of the image processing unit 112, and the face / non-face area setting unit 118 is set.
The coordinate data is transferred from the controller 109 to the controller 109 and its control is performed, and the face / non-face area separation / synthesis unit 11
The data is read from the image processing unit 112 by 1 and written to a predetermined position address of the frame memory 105.

Next, the compressed image data is transferred from the external storage device 117 to the multiplexing unit 116, the area data is transferred to the face / non-face area setting unit 118, and the image expansion / contraction is started. Taken out by the non-face area separation / synthesis unit 111,
The controller 109 transfers the coordinate data to the face / non-face area setting unit 118 so that the image data can be written at the face image position in the ID image on the frame memory 105, and the control is performed.

After the ID image is created on the frame memory 105 and read out under the control of the controller 109, the D / A converter 106 converts the ID image into an analog video signal, and then the output signal conversion section. Converted to NTSC video signal and given to the video printer 125 for ID
The card image is printed out. This makes it possible to issue an ID card.

In the above embodiment, the operator can determine the face area from the photographed image.
It is also possible to recognize the face position and automatically determine the face area.

In order to obtain such an automated processing system, the configuration as shown in FIG. 11 may be used. In the configuration of FIG. 11, the face area coordinate automatic detection unit 126 that automatically detects the area of the face image from the image and calculates and outputs the face coordinates is connected to the data path of the frame memory 105. The coordinate values are connected so that they can be input to the face / non-face area setting unit 118.

In this operation, the face portion is extracted by image processing based on the contour of the person and the skin color (Rs ± ΔR, Gs ± ΔG, Bs ± ΔB) as the complexion, and the region is centered. , Calculate the frame coordinates with a certain distance. Subsequent processing may be performed in the same manner as the above-described embodiment, and by doing so, a series of these processing can be automatically performed.

A method of recognizing the face position and automatically deciding the face region and extracting the face region will be described below. This is because after the face portion is roughly recognized, the position of the eyes in the face portion is recognized and, for example, a small area B including both eyes as shown in FIG. 12 is detected. The area coordinate detector 126 causes the calculation.

The reason why only the eyes are surrounded is that when a person looks at a person, he or she mainly looks at the eyes animally. It is also an academically well-known fact that the condition of the eyes and nose clearly shows the characteristics of the human face.

Therefore, if the resolution of the image in the vicinity of the eye is maintained without deteriorating, the perceived degree of fineness of the image received from the whole when the ID image is viewed is about the same as the resolution of the image in the vicinity of the eye. As described above, the effect of illusion by the observer can be expected, and the attention area (B area) in the image is a narrow area near the eye, and another area (A area) sacrificing resolution.
Expand the image and aim for higher compression processing for all images.

In order to detect the eye position, first, edge detection is performed when the human area is separated from the background, and only pixels of skin color (Rs ± ΔR, Gs ± ΔG, Bs ± ΔB) are detected in the area. Then, the image area is separated. Then the image 2
If the value is converted, an image with pixels missing in the eyes, nostrils, mouth, etc. can be obtained, and the white color of the white part of the original image at the pixel position where the pixels are removed is detected, and the white area becomes the skin color area. On the other hand, if the center of gravity is at the top, it is recognized as the eye position.

The above calculation is performed, and the frame coordinates as shown in FIG. 12 are calculated so as to surround the position. If the subsequent processing is performed in the same manner as in the above-described embodiment, these series of processing can be automatically performed.

In these examples, when the compression ratio is changed for each block, the connection between blocks may be noticeable due to the difference in blocks. However, as in the embodiments of the present invention, the image processing unit and the DCT processing unit are Provide and image processing D
If it is separated from CT processing, the image processing unit performs image smoothing processing on the connected portions of blocks after decompression,
You can also make it inconspicuous.

As described above, according to the present invention, the image data of the subject image is held in the image memory for temporarily holding the image data, and the image data in the image memory is divided into the set attention area and the other areas, and the predetermined pixels are respectively provided. The data is read out in block units of size, an orthogonal transformation operation such as a DCT operation is performed, and the data obtained by this transformation is subjected to predetermined quantization in accordance with the position on the image memory and the image state of the block. Quantized with a table for encoding, the quantized data is encoded and recorded together with position information on the image memory of the block and information for selection of the quantization table used, in the quantization, The image of the attention area has a lower compression rate as the image is finer,
The selection criteria of the table are defined so that the images in the other areas are compressed at a high compression rate.

Therefore, since the image of the region of interest has a low compression rate, there is no fear of deterioration even if the image is expanded, and the compression rate of other regions is high, so that the part may deteriorate when the image is expanded. Since the image quality in the attention area is ensured, by setting the attention area in the face portion like the ID image, the image quality in the necessary portion is maintained and the high data compression in other portions is performed. Due to the effect, the amount of image data can be reduced.

Therefore, it is possible to record and store the image with high compression as a whole so as not to deteriorate the sharpness of the face portion in the image,
In addition, since the compression ratio is changed even in the non-face part according to the difference in the amount of information for each part, even in the image outside the face part, it is possible to maintain a proper sharpness and maintain the image quality level that can be used for printing. It is possible to reduce the amount of image data of, and it is possible to record and save a large number of images.

The present invention is not limited to the above-described embodiment, but can be carried out by appropriately modifying it within the scope not changing the gist thereof. For example, in the embodiment, an ID face image Although the description has been centered on compression, a sufficient effect can be expected when applied not only to the compression of the ID face image but also to general image compression filing.

[0110]

As described in detail above, according to the present invention,
While maintaining the image quality level that can be used for printing, the amount of image data per sheet can be reduced, and a large number of images can be recorded and saved.

[Brief description of drawings]

FIG. 1 is a block diagram showing the overall configuration of an embodiment of the present invention.

FIG. 2 is a diagram for explaining area division used in the present invention.

FIG. 3 is a diagram for explaining an area-based compression rate weight table used in the present invention.

FIG. 4 is a diagram for explaining DCT coefficients obtained by a DCT calculation in a DCT processing unit 113 according to the embodiment of the present invention.

FIG. 5 is a view for explaining a weight table of block-by-block compression rates used in the present invention.

FIG. 6 is a diagram for explaining a relationship between a table number selected by a table selection unit used in the present invention and a W value.

FIG. 7 is a diagram for explaining another embodiment of the present invention.

FIG. 8 is a diagram for explaining another embodiment of the present invention.

FIG. 9 is a diagram for explaining another embodiment of the present invention.

FIG. 10 is a diagram for explaining another embodiment of the present invention.

FIG. 11 is a diagram for explaining another embodiment of the present invention.

FIG. 12 is a diagram for explaining another embodiment of the present invention.

FIG. 13 is a diagram for explaining a conventional technique.

FIG. 14 is a diagram for explaining a conventional technique.

[Explanation of symbols]

101 ... Subject, 102 ... Background, 103 ... Color TV
Camera, 104 ... A / D converter, 105 ... frame memory, 106 ... D / A converter, 107 ... output signal converter,
108 ... TV monitor, 109 ... Controller, 110 ...
Keyboard, 111 ... Face / non-face area separation / synthesis unit, 1
12 ... Image processing unit, 113 ... DCT processing unit, 114 ... Quantization unit, 123 ... Quantization table, 115 ... Encoding unit,
116 ... Multiplexing unit, 117 ... External storage device, 118 ... Face / non-face area setting unit, 119 ... Resolution determination unit, 120
... Block compression rate weight table, 121 ... Area compression rate weight table, 122 ... Compression rate setting and area reading section, 123 ... Quantization table, 124 ... Table selection section, 125 ... Video printer, 126 ... Face area coordinates Automatic detection unit.

Claims (1)

[Claims] 1. An image memory for temporarily holding input image data, setting means for setting an attention area of an image on the image memory and another area, and the set attention area and the other area. Separating means for dividing and reading out from the image memory in block units each having a predetermined pixel size for each area; a quantization table provided with a plurality of sets of quantization tables having different compression rates; Selection means for determining the fineness of the image of the read block, and a selection table for selecting a compression rate table determined from the determination result of the determination means and the area to which the read block belongs from the quantization table. Means, an orthogonal transformation means for orthogonally transforming the image data read by the reading means, and the data obtained by the orthogonal transformation is selected by the selection means. Quantizing means for quantizing the quantized data, coding means for coding the quantized data, position information at the time of reading the coded data by the reading means, and the table used. The image of the region of interest is compressed with a lower compression ratio as the image is finer, and the images of the other regions are compressed with a higher compression ratio. As described above, the image information compression apparatus characterized in that the selection criterion of the table is defined.