JP2006031597A - Motion vector detection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a motion vector detection device which is applied particularly to the detection of a motion vector by a gradient method using an initial deflection vector, or the like, and can detect a motion vector highly precisely corresponding to various resolutions and amounts of motion without increasing the scale of the circuit substantially. <P>SOLUTION: The device switches the selection position of the initial deflection candidate vectors VK1-VK8 according to the resolution of video signals and the amounts of motion. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a motion vector detection apparatus, and in particular, can be applied to detection of a motion vector by a gradient method using an initial deviation vector. According to the present invention, it is possible to detect a motion vector with high accuracy corresponding to various resolutions and motion amounts without significantly increasing the circuit scale by switching the selection position of the initial deviation candidate vector. To.

  Conventionally, in high-efficiency encoding processing of video signals, motion vectors are used to improve the efficiency of inter-frame encoding processing. In television signal format conversion, motion using motion vectors is used. By the correction, the jerky movement due to the conversion of the number of fields is corrected to a smooth movement.

  In such motion vector detection, a video signal is blocked into motion vector detection units (that is, macro blocks) of m pixels × n lines (m and n are integers), and a motion vector is detected for each block. For example, there are a pattern matching method disclosed in Japanese Patent Laid-Open Nos. 55-162683 and 55-162684, a gradient method disclosed in Japanese Patent Laid-Open No. 60-158786, a phase correlation method, and the like. It is made to be used.

  Among such motion vector detection methods, for the gradient method, for example, Japanese Patent Application Laid-Open No. 62-206980 proposes a method for detecting a motion vector using an initial deviation vector. Here, the initial deviation vector is a predicted value of the motion vector, and is selected from a plurality of initial deviation candidate vectors. In addition, in each initial deviation candidate vector, a macroblock (hereinafter, referred to as a macroblock) associated with a predetermined selection position around a macroblock (hereinafter referred to as a detected block) relating to the detection of a motion vector from the already detected motion vector. The motion vector already detected at the selected position) is applied. In this motion vector detection method using an initial displacement vector, Japanese Patent Laid-Open No. 4-78286 discloses a method of detecting a displacement vector with reference to the initial displacement vector, and then the initial displacement vector and the displacement vector. A method of detecting a motion vector with a simple configuration by detecting a motion vector by addition to the above is disclosed. According to this detection method, a motion vector can be detected by motion vector = initial displacement vector + deviation vector. For example, when the motion is uniform such as camera pan, the displacement vector is “0”. Thus, only the initial deviation vector has a value.

  However, the method of detecting a motion vector by the gradient method is a method of detecting a motion vector by an arithmetic expression using an image gradient and an inter-frame difference value, and has high detection accuracy but only detection accuracy for several pixels. In the region where the amount of motion changes suddenly, the detection accuracy deteriorates, and there is a problem that it takes several fields to follow the true amount of motion.

  As one method for solving this problem, if many selection positions are set in the detected block and the number of initial deviation candidate vectors is increased, the initial deviation vector can be detected following the movement, Thus, it is considered that a motion vector can be detected following a sudden change in motion.

However, if the number of initial deviation candidate vectors is increased, there is a problem that the circuit scale related to selection of the initial deviation vector increases accordingly. In particular, in the case of detecting a motion vector in real time, it is necessary to select an initial deviation candidate vector by parallel processing, which inevitably increases the circuit scale.
Japanese Unexamined Patent Publication No. 55-162683 Japanese Patent Laid-Open No. 55-162684 JP 60-158786 A JP-A-4-78286

  The present invention has been made in consideration of the above points, and motion vector detection capable of detecting a motion vector with high accuracy corresponding to various resolutions and motion amounts without significantly increasing the circuit scale. The device is to be proposed.

  In order to solve such a problem, the invention according to claim 1 selects an initial deviation vector from a plurality of initial deviation candidate vectors, detects the deviation vector using the initial deviation vector, and corrects the initial deviation vector. By applying to a motion vector detection device that detects motion vectors of video signals that are sequentially input, a plurality of initial deviation candidate vectors are selected from already detected motion vectors at a predetermined selected position around the detected block. And a selection position switching means for switching the selection position of the initial deviation candidate vector by the selection means.

  According to a second aspect of the present invention, in the configuration of the first aspect, the selection position switching means switches the selection position in accordance with the resolution and / or motion amount of the video signal.

  According to a third aspect of the present invention, in the configuration of the second aspect, the selected position switching means moves the selected position away from the detected block as the resolution of the video signal becomes higher.

  According to a fourth aspect of the present invention, in the configuration of the second aspect, the selected position switching means moves the selected position away from the detected block as the movement of the video signal increases.

  According to the configuration of claim 1, the initial deviation vector is selected from a plurality of initial deviation candidate vectors, the deviation vector is detected using the initial deviation vector, and the initial deviation vector is corrected to be sequentially input. Applying to a motion vector detecting device for detecting a motion vector of a video signal to be detected, selecting means for selecting a plurality of initial deviation candidate vectors from already detected motion vectors at a predetermined selected position around the detected block, and selecting means And a selection position switching means for switching the selection position of the initial deviation candidate vector by the initial deviation candidate reflecting the movement at the selection position away from the detected block by switching the selection position when the amount of movement is large. Even if the vector can be selected and the amount of motion is the same, if the resolution is high, the selected position is switched and the detected block is changed. Tsu and away initial excursion candidate vector that reflects the amount of movement at a selected position from the click can be selected. As a result, it is possible to select the initial deviation candidate vector reflecting the movement by switching the selection position without selecting many initial deviation candidate vectors, thereby variously without greatly increasing the circuit scale. The motion vector can be detected with high accuracy corresponding to the resolution and the motion amount.

  According to the configuration of claim 2, in the configuration of claim 1, the selection position switching means switches the selection position according to the resolution and / or the amount of motion of the video signal, so that the selected position is changed according to the input video signal. A highly accurate motion vector can be easily detected.

  According to the configuration of claim 3, in the configuration of claim 2, the selection position switching means moves the selection position away from the detected block as the resolution of the video signal becomes higher, so that the initial deviation candidate when the resolution changes. Degradation of vector accuracy can be avoided.

  According to the configuration of claim 4, in the configuration of claim 2, the selected position switching means moves the selected position away from the detected block as the motion of the video signal increases, so that an appropriate initial displacement according to the amount of motion is achieved. Candidate vectors can be detected.

  According to the present invention, in motion vector detection by a gradient method or the like using an initial deviation vector, a motion vector can be obtained with high accuracy corresponding to various resolutions and motion amounts without significantly increasing the circuit scale. Can be detected.

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

(1) Configuration of Embodiment FIG. 1 is a block diagram showing a motion vector detection apparatus according to an embodiment of the present invention. This motion vector detection device 1 is applied to, for example, a format conversion device, a coding device based on MPEG (Moving Picture Experts Group), a noise reduction device, etc., and detects a motion vector V from a luminance signal Y constituting a video signal. Output. In the detection of the motion vector, not only when the luminance signal Y is used, but the color difference signals RY and BY can also be used.

  The motion vector detection device 1 detects an approximate motion vector from the initial displacement vector V1, detects the displacement vector ΔV by a gradient method using the detected motion vector, and corrects the initial displacement vector V1. Thus, the motion vector V is detected.

  That is, in the motion vector detection device 1, the pre-filter 2 is a two-dimensional low-pass filter, which receives the luminance signal Y and suppresses high frequency components, thereby preventing erroneous detection of motion vectors due to noise. In addition, when the luminance signal Y is based on an interlaced video signal, the pre-filter 2 executes so-called center-of-gravity correction processing, so that when detecting a motion vector between consecutive fields, the lines differ between the fields. This prevents the deterioration of the motion vector detection accuracy. The pre-filter 2 is not limited to a low-pass filter, and a band-pass filter can be applied.

  The initial deviation vector selection circuit 3 inputs initial deviation candidate vectors VK1 to VK8 from the motion vectors V of the previous field and the current field that are already detected and held in the memory circuit 6, and these initial deviation candidates. The most appropriate vector VK1 to VK8 is selected and output as the initial displacement vector V1.

  The gradient method computing circuit 4 detects the displacement vector ΔV by the gradient method with reference to the initial displacement vector V1 output from the initial displacement vector selection circuit 3. That is, the gradient method arithmetic circuit 4 has a one-field delay circuit that outputs the luminance signal Y output from the pre-filter 2 with a delay of one field, and the luminance signal output from the one-field delay circuit is initialized. The motion vector is corrected by the position vector V1, and the displacement vector ΔV is detected by the calculation by the gradient method between the motion-corrected luminance signal and the luminance signal Y output from the prefilter 2.

  The addition circuit 5 corrects the initial displacement vector V1 by adding the displacement vector ΔV to the initial displacement vector V1, and outputs the result as a motion vector V. The memory circuit 6 records and holds the motion vector V output from the adder circuit 5. Thus, the motion vector detection apparatus 1 selects a plurality of initial deviation candidate vectors VK1 to VK8 from the already detected motion vectors, and detects a motion vector V from the plurality of initial deviation candidate vectors VK1 to VK8. A suitable initial deviation candidate vector is selected and set as the initial deviation vector V1, and the motion vector V is detected based on the initial deviation vector V1.

  In this series of processing, the initial deviation vector selection circuit 3 selects the already detected motion vector held in the memory circuit 6 at a predetermined selection position set around the detected block, and the initial deviation candidate vector VK1. ˜VK8 is input, and the selected position is switched by the operation of the operator.

  Here, in this embodiment, the relationship between the detected block and the macroblock related to such a selected position is defined by four types of patterns as shown in FIGS.

  Here, the first pattern is a pattern corresponding to standard image quality and low-speed movement, and as shown in FIG. 2A, the surrounding eight macroblocks MB1 adjacent to the detected block MB0 for detecting the motion vector V are used. ... MB8 is selected as the selected position. As a result, when the first pattern is selected by the operator's operation, the initial deviation vector selection circuit 3 reads out the motion vectors already detected in these macro blocks MB1 to MB8 from the memory circuit 6 to obtain initial deviation candidates. The vectors VK1 to VK8 are set.

  As a result, for an image with standard image quality and slow motion, a motion vector already detected at a position close to the detected block MB0 can be selected as the initial deviation candidate vectors VK1 to VK8. .

  On the other hand, the second pattern is a pattern corresponding to standard image quality and high-speed movement, and as shown in FIG. 2B, two macroblocks MB3 adjacent in the vertical direction to the detected block MB0, MB5, macroblocks MB2, MB7 and MB4, MB8 on the left and right of these macroblocks MB3 and MB5, and macroblocks MB1 and MB6 adjacent to the left and right of detected block MB0 are selected positions. As a result, when the second pattern is selected by the operator's operation, the initial deviation vector selection circuit 3 reads the motion vectors already detected in these macroblocks MB1 to MB8 from the memory circuit 6 to obtain initial deviation candidates. The vectors VK1 to VK8 are set.

  As a result, for an image with standard image quality and fast motion, the motion vector already detected at a position farther to the left and right of the detected block MB0 than the standard image quality / low-speed movement is used as an initial deviation candidate. The vectors VK1 to VK8 can be selected.

  The third pattern is a pattern corresponding to high-definition image quality and low-speed movement. As shown in FIG. 3A, two macroblocks MB3 and MB5 that are adjacent to the detected block MB0 in the vertical direction are adjacent to each other. And two macroblocks MB1 and MB6 that are adjacent to the detected block MB0 in the left-right direction and four macroblocks MB2, MB4, MB7, and MB8 that are adjacent to the detected block MB0 in an oblique direction are selected. It is made to be a position. As a result, when the third pattern is selected by the operator's operation, the initial deviation vector selection circuit 3 reads out the motion vectors already detected in these macro blocks MB1 to MB8 from the memory circuit 6 to obtain initial deviation candidates. The vectors VK1 to VK8 are set.

  As a result, an image with high definition image quality and slow motion has already been detected at a position far from the detected block MB0 on the basis of the number of pixels as compared with the case of standard image quality / low-speed movement. The selected motion vectors can be selected as initial deviation candidate vectors VK1 to VK8.

  On the other hand, the fourth pattern is a pattern corresponding to high-definition image quality and high-speed movement. As shown in FIG. 3B, two macros that are adjacent to the detected block MB0 in the vertical direction are adjacent to each other. Blocks MB3 and MB5, and macroblocks MB1 and MB6 that are adjacent to each other with respect to the detected block MB0 in the left-right direction, and macroblocks that are adjacent to the detected block MB0 in the horizontal direction Further, four macroblocks MB2, MB4, MB7, and MB8 adjacent in the oblique direction are selected positions. As a result, when the fourth pattern is selected by the operator's operation, the initial deviation vector selection circuit 3 reads out the motion vectors already detected in these macroblocks MB1 to MB8 from the memory circuit 6 to obtain initial deviation candidates. The vectors VK1 to VK8 are set.

  As a result, for high-definition image quality and fast motion, the motion vectors already detected at positions farther to the left and right of the detected block MB0 than the case of high-definition image quality / low-speed movement are initialized. The deviation candidate vectors VK1 to VK8 can be selected. Incidentally, such a selection position is set at a position away from the detected block MB0 as compared with the case of standard image quality and high-speed movement.

  As described above, the initial displacement vector selection circuit 3 prepares four patterns related to the macroblocks MB1 to MB8 for selecting the initial displacement candidate vectors VK1 to VK8 according to the resolution and motion of the input video signal. These patterns are selected by a switching control signal S1 output from the operation unit.

  In this embodiment, among the macroblocks MB1 to MB8, the motion vectors already detected in the current field are applied to the macroblocks MB1 to MB3, and the macroblocks MB4 to MB8 are already detected in the current field or the previous field. However, the present invention is not limited to this. For example, the motion vectors detected in the previous field may be applied to all macroblocks MB1 to MB8. Various patterns can be applied to which of the motion vectors detected in the current field or the previous field for MB1 to MB8.

  The initial deviation vector selection circuit 3 reads out the motion vectors already detected in the macroblocks MB1 to MB8 according to the pattern thus selected from the memory circuit 6 and sets them as initial deviation candidate vectors VK1 to VK8. An optimum initial deviation vector V1 is detected from these.

  That is, the initial deviation vector selection circuit 3 records and holds the luminance signal Y related to the current field output from the pre-filter 2 in an internal one-field delay circuit (not shown), and delays it for a period of one field. Thus, the luminance signal Y1 relating to the previous field is output. The initial displacement vector selection circuit 3 is an initial unit for each write block used for writing the luminance signal Y1 related to the previous field in units of macroblocks of the luminance signal Y related to the current field output from the prefilter 2. The luminance signal Y1 held by the address control in which the read address is shifted is read by the deviation candidate vectors VK1 to VK8, and the luminance signal Y1 held thereby is corrected by the initial deviation candidate vectors VK1 to VK8. Further, between the read luminance signal Y1 and the luminance signal Y output from the prefilter 2, an inter-field difference value is calculated for each of the initial deviation candidate vectors VK1 to VK8, and the inter-field difference value is calculated. A small initial deviation candidate vector is selected and output as an initial deviation vector V1. In addition, in the inter-field difference values relating to the initial deviation candidate vectors VK1 to VK8, although the sum of absolute differences is applied, the sum of squares of the pixel values may be applied.

  Accordingly, in this embodiment, the initial deviation vector selection circuit 3 selects a plurality of initial deviation candidate vectors from the already detected motion vectors at a predetermined selection position around the detected block, and an initial by the selection means. Selection position switching means for switching the selection position of the deviation candidate vector is configured.

(2) Operation of Embodiment In the above configuration, in the motion vector detection apparatus 1, the luminance signal Y of the video signal to be processed is corrected for the frequency characteristics by the prefilter 2, and then the initial displacement vector is selected. The circuit 3 detects the initial displacement vector V1, and the displacement vector ΔV is detected by the gradient method processing in the gradient method computing circuit 4 using the initial displacement vector V1. In the luminance signal Y, the displacement vector ΔV is added to the initial displacement vector V1 to detect the motion vector V, and the motion vector V is recorded in the memory circuit 6 and applied to, for example, field interpolation. .

  In addition, the motion vector V recorded in the memory circuit 6 corresponds to the subsequent motion vector detection target macroblock, and a plurality of motion vectors from the memory circuit 6 are used as the initial displacement candidate vectors VK1 to VK8. The initial deviation vector selection circuit 3 detects the inter-field difference value for each of the initial deviation candidate vectors VK1 to VK8, and the motion vector having the smallest inter-field difference value is the initial deviation vector V1. Selected as.

  As a result, the motion vector detection apparatus 1 can detect the motion vector V with high detection accuracy by effectively using the correlation between successive fields. That is, for example, when the movement such as camera pan is constant, initial deviation candidate vectors VK1 to VK8 are detected by the same value reflecting this movement in almost all macroblocks, and thereby the initial deviation vector reflecting the movement. The displacement vector of the value “0” is detected by V1, and the motion vector is detected with high accuracy. Further, for example, in a macroblock related to an object moving in the screen, if the motion vector V of this object has already been detected in any of the initial deviation candidate vectors VK1 to VK8, the motion vector of this object An initial deviation candidate vector based on a macroblock from which V has been detected is selected and set as the initial deviation vector V1, and in this case as well, the motion vector V is detected with high accuracy.

  However, when these initial deviation candidate vectors VK1 to VK8 do not include a motion vector relating to an object moving in the screen, the initial deviation vector V1 having a greatly different value is used in the detection of the initial deviation vector V1. By executing with reference, the initial displacement vector V1 is erroneously detected, and in this case, the motion vector V of this object cannot be detected correctly. As a result, in this case, the motion vector V cannot be detected with good follow-up in a region where the motion related to the object changes suddenly.

  Even in such a case, if the number of initial deviation candidate vectors is increased, there is a high possibility that a motion vector of such a motion object is detected in any of the initial deviation candidate vectors. As a result, the motion vector V can be detected with good followability with respect to such a sudden change in motion related to the object. However, in this case, the circuit scale of the motion vector detection device greatly increases as the number of initial deviation candidate vectors increases.

  For this reason, in this embodiment, in the selection of the initial deviation candidate vectors VK1 to VK8 in the initial deviation vector selection circuit 3, the selection positions of the initial deviation candidate vectors VK1 to VK8 can be switched. When the followability deteriorates due to the above, the selection position is switched so that the followability is improved, and the motion vector detection accuracy is improved.

  That is, in the region where the motion changes, if the detection accuracy of the motion vector V is inferior, it can be said that the initial displacement candidate vectors VK1 to VK8 related to the detection of the motion vector V do not reflect the motion related to this region. In this case, by switching the selection positions of the initial deviation candidate vectors VK1 to VK8, the already detected motion vector V related to this region can be included in the initial deviation candidate vectors VK1 to VK8. It is thought that the followability can be improved by improving the detection accuracy.

  Specifically, when the motion is small, there is a high possibility that the motion vector of the object related to this motion is detected at the selected position near the detected block MB0. On the other hand, when the motion is large, it is less likely that the motion vector of the object related to the motion is detected at the selected position near the detected block MB0. There is a high possibility that the motion vector of the object related to this motion is detected at the selected position away from the detected block MB0.

  As a result, the initial displacement vector selection circuit 3 switches the selection position so as to move away from the detected block MB0 as the movement increases. Thereby, the detection accuracy of the motion vector V is improved.

  Even if the amount of motion is the same, if the resolution is increased, the motion vector of the object related to the amount of motion is detected at the selected position far from the detected block MB0, so that in this case as well, the initial deviation is detected. The selection position is switched by the vector selection circuit 3 so as to move away from the detected block MB0 as the resolution increases. Thereby, the detection accuracy of the motion vector V is improved.

(3) Advantages of the embodiment According to the above configuration, the circuit scale is greatly increased in detecting a motion vector by a gradient method using the initial deviation vector by switching the selection position of the initial deviation candidate vector. Therefore, it is possible to detect a motion vector with high accuracy corresponding to various resolutions and motion amounts.

  Further, by switching the selection position of the initial deviation candidate vector according to the resolution and / or motion amount of the video signal, it is possible to easily detect the motion vector with high accuracy according to the input video signal.

  In the above-described embodiment, the case where eight types of initial deviation candidate vectors VK1 to VK8 are selected from the periphery of the detected block MB0 has been described. However, the present invention is not limited to this, and the initial deviation is selected as necessary. The number of candidate vectors VK1 to VK8 can be set variously.

  In the above-described embodiment, the initial deviation candidate vectors VK1 to VK8 relating to the standard image quality / high speed image or the high image quality / high speed image with respect to the standard image quality / low speed image or the high definition image quality / low speed image are stored. Although the case where the selected position is widened to the left and right has been described, the present invention is not limited to this.

  In the above-described embodiment, the case where the selection position of the initial deviation candidate vectors VK1 to VK8 is switched according to the resolution by operating the operation unit has been described. However, the present invention is not limited to this, and the input video signal You may make it switch automatically according to the resolution. In this case, detection means for detecting the number of pixels in the horizontal direction and the number of lines in the vertical direction is provided, and the initial displacement vector selection circuit 3 moves the selected position from the detected block MB0 as the number of pixels and the number of lines increase as a result of this detection. If they are kept away from each other, it is possible to avoid deterioration of the accuracy of the initial deviation candidate vectors VK1 to VK8 due to different resolutions.

  In the above-described embodiment, the case where the selection position of the initial deviation candidate vectors VK1 to VK8 is switched according to the amount of motion by operating the operation unit has been described. However, the present invention is not limited to this, and the input video signal is not limited thereto. It may be switched automatically according to the amount of movement. In this case, a detecting means for detecting the motion amount based on the inter-field difference value or inter-frame difference value of the video signal is provided, and the initial displacement vector selection circuit 3 selects the selected position as the motion amount increases according to these detection results. By moving away from the detected block MB0, appropriate initial deviation candidate vectors VK1 to VK8 can be detected according to the amount of motion.

  In the above-described embodiment, the case where the motion vector is detected between the fields is described. However, the present invention is not limited to this, and the motion vector may be detected between the frames. In this case, in the initial deviation vector selection circuit 3, a 1-frame delay circuit may be applied instead of the 1-field delay circuit so that the luminance signal Y is delayed by 1 frame.

  In the above-described embodiments, the case where the deviation vector is detected by the gradient method has been described. However, the present invention is not limited to this, and the deviation vector is detected by another method such as a phase correlation method. Also good.

  The present invention relates to a motion vector detection device, and can be applied to, for example, a format conversion device.

It is a block diagram which shows the motion vector detection apparatus which concerns on the Example of this invention. FIG. 3 is a schematic diagram showing a detection pattern of initial deviation candidate vectors in the initial deviation vector selection circuit of FIG. 1. FIG. 10 is a schematic diagram illustrating another example of an initial deviation candidate vector detection pattern in the initial deviation vector selection circuit of FIG. 1.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Motion vector detection apparatus, 2 ... Pre-filter, 3 ... Initial deviation vector selection circuit, 4 ... Gradient method arithmetic circuit, 5 ... Adder circuit, 6 ... Memory circuit

Claims (4)

Motion of a video signal sequentially input by selecting an initial deviation vector from a plurality of initial deviation candidate vectors, detecting the deviation vector using the initial deviation vector, and correcting the initial deviation vector In a motion vector detection device for detecting a vector,
Selecting means for selecting the plurality of initial deviation candidate vectors from already detected motion vectors at a predetermined selection position around the detected block;
And a selection position switching means for switching the selection position of the initial deviation candidate vector by the selection means. The selected position switching means is
The motion vector detection device according to claim 1, wherein the selection position is switched according to a resolution and / or a motion amount of the video signal. The selected position switching means is
The motion vector detection device according to claim 2, wherein the selected position is further away from the detected block as the resolution of the video signal is higher. The selected position switching means is
The motion vector detection device according to claim 2, wherein the selected position is moved away from the detected block as the motion of the video signal increases.

Images