KR0154948B1 - Motion vector search apparatus using the gradient and the intensity similarity of the moving picture - Google Patents

Abstract

The motion vector search apparatus using the gradient and intensity of an image according to the present invention includes first to fourth memory units 210, 215, 260, and 265, an intensity matcher 220, and a gradient calculator. Including the 250, the gradient matching unit 270, and the comparison determination unit 280, the motion vector of the video can be found more accurately.

Description

Motion vector search device using similarity of image gradient and intensity

FIG. 1 is an exemplary diagram illustrating a change in intensity between a past image and a current image.

2 is a block diagram for explaining the operation of the motion vector search apparatus using the gradient and intensity of the image according to the present invention.

* Explanation of symbols for main parts of the drawings

210, 215, 260, 265: memory 220: intensity matching

250: gradient calculation unit 270: gradient matching unit

280: comparison judgment

The present invention relates to a motion vector search device using the similarity of the gradient and intensity of an image.

In a high definition video system based on digital data transmission such as high definition television (HDTV) and multimedia, a large amount of data to be processed occurs.

However, the bandwidth available for data transmission is limited, so data must be compressed and coded in order to efficiently transmit a large amount of data. Accordingly, data encoding and decoding of encoded data are fundamental elements of a high quality video system, and research for this has been actively conducted in recent years.

In the conventional video signal coding, a video signal is processed in blocks, such as discrete cosine transform (DCT), quantization (Q), variable length coding (VLC), and the like. Encoded by the process. Here, the block is composed of NXN pixels, where N is any positive integer.

However, in the block-based processing method, a blocking effect in which a boundary between blocks is prominent due to a slight difference in brightness between adjacent blocks occurs, and this blocking phenomenon occurs particularly in low bitrate coding. In other words, the higher the compression ratio, the greater the degree of deterioration in image quality than any other noise.

Proposed to solve this blocking phenomenon is image coding using a segmentation technique, which will be described in more detail as follows.

First, an arbitrary reference value is set, and only pixels having similar characteristics based on a predetermined reference value among pixels that are the basic units of an image signal are collected and divided into one region. Texture information indicating video signal characteristics and contour information indicating a boundary between neighboring other regions are detected and coded, respectively.

On the other hand, in order to effectively compress the video signal, it is necessary to remove the correlation of the image existing in time as already known. A commonly used method for this is the inter-frame motion compensation prediction encoding method.

Here, motion compensation refers to a motion vector (MV) indicating a direction in which direction the image of the current frame has moved compared to the previous frame by a predetermined algorithm, and the image of the previous frame. To move by the motion vector. Accordingly, inter-frame motion compensation prediction encoding performs motion compensation on a previous frame using a motion vector, and compresses and encodes the difference signal between the video signal of the current frame and the motion compensated video signal.

However, since the conventional block matching algorithm uses a discrete cosine transform (DCT), a motion vector is obtained using the smallest difference value. Since this is a method of selecting only the smallest difference value rather than considering the continuity or similarity of the whole object, there has been a problem that an error may occur when there is a change in brightness between the current image and the past image. This will be described in more detail as follows.

1 (a) and (b) are exemplary diagrams illustrating changes in intensity between a past image and a current image. As shown in FIG. 1 (a), the region A and intensity B of intensity 100 of the past image frame are respectively regions A ¹ and intensity B ¹ of intensity 130 and 80, respectively, as shown in FIG. 1 (b). Was changed. As a result, there is a brightness change of +30 throughout the image.

Here, let's consider area C, which is the boundary between area A and area B. C region is substantially in the C ¹ has been moved to an area would When considering only the difference value of the picture signal the motion vector at all eongttunghage makin found to 0, which is the same with no movement than the difference value in the moved C ¹ region located This is because the difference is smaller.

Accordingly, an object of the present invention is to provide an apparatus capable of finding motion vectors more accurately by using similarity of gradient and intensity of an image.

The motion vector search apparatus using the similarity of the gradient and the intensity of the image according to the present invention for achieving the above object, the current image data (I _C ) and the past immediately preceding the current image And first and second memory units for recording the image data I _P , respectively, wherein the first memory unit has image data I at positions x + i and y + j corresponding to the input first address signal. Outputs _C (x + i, y + j), and the second memory unit has image data I _P (at position (x + i + dx, y + j + dy) corresponding to the input second address signal. x + i + dx, y + j + dy)) and a unit gradient of the current image data I _C and the past image data I _P (Unit Gradient: U _C and U _P). ) And third and fourth memory sections respectively recording the third memory section, and the third memory section corresponds to a first address signal input thereto. (x + i, y + j) units gradient _{(U C (x + i,} y + j)) output, and wherein the fourth memory unit corresponding to a second address signal inputted position (x + i + dx of and a gradient memory unit for outputting a unit gradient (U _P (x + i + dx, y + j + dy)) of y + j + dy, and calculating a unit gradient of input image data. A gradient calculator which outputs to a gradient memory unit and symmetrically outputs to the third and fourth memory units on a frame-by-frame basis; and image data I to find a current motion vector in response to a clock signal CLK. Image data and unit gradient present in the average value area B centering on the unit gradient U _C (x, y) corresponding to _C (x, y)) and the image data I _C (x, y) Outputs a first address for specifying the position (x + i, y + j) of the image data and the unit gradient in the motion vector search range (S). Output a second address for specifying the position of (x + i + dx, y + j + dy) and output the shifted displacements (dx, dy) for the motion vector search, but within the average value area (B) The first control signal C1 is output when the positioning of all the image data and the unit gradient is completed, and the second control signal C2 when the positioning of all the image data and the gradient in the motion vector search region S is completed. ) And the image data (I _C (x + y, y + j) and (x + i + dx, y + j + dy)) are input from the first and second memory units. Thus intensity error values to the intensity matching formula _{(E I (x, y)} ) the calculation and intensity matching in response to the first control signal (C1) output the intensity error value _{(E I (x, y)} ) And a predetermined gradient matching equation by receiving unit gradients U _C (x + i, y + u) and p (x + i + dx, y + j + dy) from the third and fourth memory units. Depending on yeah Gradient error value _{(E V (x, y)} ) the calculation and, in response to the first control signal (C1) the gradient error value _{(E V (x, y)} ) for the gradient matching unit for outputting the motion vector (M _X , M _Y ) and a determination value are recorded, and the intensity error value E _I (x, y) and the gradient error value E _V (x, from the intensity matching unit and the gradient matching unit, respectively, are recorded. y)), and the result value E (x, y) obtained by performing a predetermined decision expression is compared with the decision value, and the result of comparison (E (x, y)) is greater than the decision value. If small, the result value E (x, y) is recorded as a new determination value and the displacement (dx, dy) moved for the motion vector search provided from the controller is recorded as a new motion vector, comparison determination unit when the second control signal (C2) is input is output to the motion vector (M _X, M _Y) that the current record and a reset It is configured.

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

FIG. 2 is a block diagram schematically illustrating an operation of a motion vector search device using similarity between gradients and intensities of an image according to the present invention. (210, 215, 260, 265), the gradient calculator 250, the intensity matching unit 220, the gradient matching unit 270, the comparison determination unit 280, and the controller (shown) No).

The first and second memory units 210 and 215 respectively record the current image data I _C and the past image data I _P immediately preceding the current image, wherein the image data is the intensity of each pixel. Intensity). In this case, the first memory unit 210 is the image data I _C (x + i, y +) of the position (x + i, y + j) corresponding to the first address signal provided from the controller (not shown) to be described later. j)), and the second memory unit 215 stores the image data I _{P at} the position (x + i + dx, y + j + dy) corresponding to the second address signal provided from the controller (not shown). (x + i + dx, y + j + dy))

The third and fourth memory units 260 and 265 are unit gradients of current image data I _C and past image data I _P recorded in the first and second memory units 210 and 215, respectively. Record U _C and U _P , respectively. The third memory unit 260 may have a unit gradient U _C (x + i, y + j) at a position (x + i, y + j) corresponding to the first address signal provided from a controller (not shown) to be described later. ), And the fourth memory unit 265 has a unit gradient U _P (x) of the position (x + i + dx, y + j + dy) corresponding to the second address signal provided from the controller (not shown). + i, y + j))

The gradient calculator 250 calculates and outputs a unit gradient of the input image data, and symmetrically outputs the third and fourth memory units 260 and 265 in units of frames.

The controller (not shown), in response to the clock signal CLK, the image data I _C (x, y) to find the current motion vector and the unit gradient U _C (x) corresponding to the image data. outputs a first address for designating the image data existing in the average value area B and the position of the gradient (x + i, y + j) with respect to (y)) and exists in the motion vector search range (S). The second address for designating the position (x + i + dx, y + j + dy) of the image data and the unit gradient is output, and the shifted displacements (dx, dy) are output for the motion vector search. In this case, the controller (not shown) outputs the first control signal C1 when all image data in the average value area B and the unit gradient are completed, and all image data in the motion vector search area S. When the positioning of the unit gradient is completed, the second control signal C2 is output.

Here, the intensity of each pixel used in the intensity matching of the present invention means an intensity average value of pixels existing in the average value area B around each pixel. In addition, although the size of the average value area B may vary, the exemplary embodiment is a 5 × 5 pixel block, and thus, the average value area B is positioned at the center of the pixel (x, y). It is a range (x-2 ≦ B _X ≦ x + 2, y-2 ≦ B _Y ≦ y + 2) to be moved up, down, left, and right by 2 pixels.

Intensity matching unit 220 is the intensity of the current image and the past image from the first and second memory unit (210 and 215) I _C (x + i, y + j) and I _P (x + i + dx, y + J + dy)) and calculates the intensity error value E _I (x, y) according to a preset intensity matching equation, and in response to the first control signal C1, the intensity error value E _I (x , y))

The gradient matching unit 270 is a unit gradient (U _C (x + i, y + j) and U _P (x + i + dx, the current image and the past image from the third and fourth memory unit 260 and 265) y + j + dy)) and calculates the gradient error value E _V (x, y) according to the preset gradient matching equation, and responds to the gradient error value E _V (in response to the first control signal C1). x, y)) At this time, since the direction vector is more effective in considering the image shape than the size of the gradient, in the present invention, a gradient having a size of 1, that is, a unit gradient, is used, and a calculation process for obtaining a unit gradient is well known. The description thereof will be omitted.

The comparison determiner 280 receives the intensity error value E _I (x, y) and the gradient error value E _V (x, y) from the intensity matcher 220 and the gradient matcher 270, respectively. The result value E (x, y) obtained by performing the predetermined judgment expression is compared with the preceding result value. As a result of the comparison, if the current result is smaller than the previous result, the current result is newly recorded, and the shifted displacement (dx, dy) is newly recorded for the motion vector search provided from the controller (not shown). . While repeating this process, the comparison determination unit 280 responds to the second control signal C2 provided from the control unit (not shown) to convert the currently recorded displacements dx and dy into the motion vectors M _X ,. M _Y ) as output and reset.

The operation of the motion vector search apparatus using the similarity between the gradient and the intensity of the image according to the present invention will be described in detail.

First, the first and third memory units 210 and 260 each have an intensity (I _C (x + i, y + j)) of an image corresponding to a first address signal provided from a controller (not shown). And the unit gradient U _C (x + i, y + j) are output to the intensity matching unit 220 and the gradient matching unit 270, respectively.

On the other hand, the second and fourth memory portions (215 and 265), respectively, control the second image intensity (I _P (x + i, y + j)) of the location corresponding to the address signal supplied from the (not shown) And the unit gradient U _P (x + i, y + j) are output to the intensity matching unit 220 and the gradient matching unit 270, respectively.

Then, the intensity matching unit 220 obtains the intensity I _C (x + i, y + j) and I _C (x + i, y + j) from the first and second memory units 210 and 215, respectively. The intensity error value E _I (x, y) between the average value area B of the present and past image frames is calculated according to the following equation (1).

The intensity matching unit 220 compares the intensity error value E _I (x, y) obtained as described above in response to the first control signal C1 provided from the control unit (not shown). Will output

On the other hand, the gradient matching unit 270 is a unit gradient (U _C (x + i, y + j) and U _P (x + i + dx, y + j +) from the third and fourth memory unit (260 and 265) dy)), respectively, and calculates a gradient error value E _V (x, y) between the average value region B of the current and past image frames according to Equation (2) below.

At this time, the intensity error value uses the intensity difference between the two mean value areas B as shown in equation (1), while the gradient error value is the unit gradient of the two mean value areas B as shown in equation (2). Use the inner product

The gradient matching unit 270 compares the gradient error value E _V (x, y) obtained as described above in response to the first control signal C1 provided from the control unit (not shown). Will output

Subsequently, the comparison determining unit 280 provides an intensity error value E _I (x, y) and a gradient error value E _V (x, y) from the intensity matching unit 220 and the gradient matching unit 270. It receives and performs following formula (3).

Here, W ₁ and W ₂ are values that can be appropriately selected as weights. The smaller the result value (E (x, y)) of Equation (3), the more preferable matching is achieved. As shown in Equation (3), the smaller the intensity error value E _I (x, y) is, As the gradient error value E _V (x, y) is closer to 1, more accurate matching is performed.

Accordingly, the comparison determination unit 280 compares the currently obtained result value E (x, y) with the preceding result value and the current result value E (x, y) is smaller than the preceding result value. In addition, the newly obtained result (E (x, y)) and the shifted displacement (dx, dy) for the search of the motion vector provided from the controller (not shown) are newly recorded, and vice versa. And maintain the displaced displacement for the movement search.

While this process is repeated continuously, the displacements (dx, dy) recorded in the comparison judgment unit 280 are the displacements when the smallest result value E (x, y) is obtained until now, and ultimately the best It can be said that it is obtained in a state where a match is made.

At this time, when the second control signal C2 is input, the comparison determination unit 280 outputs the displacements dx and dy currently recorded as the motion vectors M _X and M _Y.

Through the above-described process, the motion vector search apparatus using the similarity between the gradient and the intensity of the image according to the present invention has a great advantage of finding the motion vector of the video more accurately by sufficiently considering the change in brightness of the image.

Claims (4)

And first and second memory units for recording current image data I _C and past image data I _P immediately preceding the current image, respectively, wherein the first memory unit corresponds to an input first address signal. Outputs the image data I _C (x + i, y + j) of the position (x + i, y + j), and the second memory unit (x + i) corresponding to the input second address signal. an image memory section for outputting the image data I _P (x + i + dx, y + j + dy) of + dx, y + j + dy; And a third and fourth memory unit for recording unit gradients (U _C and U _P ) of the current image data I _C and the past image data I _P , respectively. Outputs a unit gradient U _C (x + i, y + j) of a position (x + i, y + j) corresponding to the input first address signal, and the fourth memory unit inputs the second address signal A gradient memory unit for outputting a unit gradient U _P (x + i + dx, y + j + dy) at a position corresponding to (x + i + dx, y + j + dy); A gradient calculator which calculates a unit gradient of the input image data and outputs the unit gradient to the gradient memory unit, and symmetrically outputs the frame to the third and fourth memory units; In response to the clock signal CLK, the image data I _C (x, y) for which the current motion vector is to be found and the unit gradient U _C (x, y) corresponding to the image data are centered. Outputs a first address for designating the image data existing in the average value area B and the position (x + i, y + j) of the unit gradient, and the image data and unit existing in the motion vector search range S. Output a second address for specifying the position of the gradient (x + i + dx, y + j + dy), and outputs the shifted displacements (dx, dy) for the motion vector search, the average value area (B) The first control signal C1 is output when the positioning of all the image data and the unit gradient is completed, and the second control signal when the positioning of all the image data and the gradient in the motion vector search region S is completed. A control unit for outputting C2); Image data I _C (x + i, y + j) and IP (x + i + dx, y + j + dy) are input from the first and second memory units, and the intensity is in accordance with a preset intensity matching equation. the error value _{(E I (x, y)} ) the calculation and intensity matching unit in response to the first control signal (C1) output the intensity error value _{(E I (x, y)} ) and; Unit gradients U _C (x + i, y + j) and U _P (x + i + dx, y + j + dy) are received from the third and fourth memory units according to a preset gradient matching equation. gradient error value _{(E V (x, y)} ) the calculation and the gradient matching unit for outputting the gradient error value _{(E V (x, y)} ) in response to the first control signal (C1) and; The motion vector (Mx, My) and the determination value are recorded, and the intensity error value E _I (x, y) and the gradient error value E _V (x, from the intensity matching unit and the gradient matching unit, respectively, are recorded. y)) and the result value (E (x, y)) obtained by performing a predetermined decision expression is compared with the decision value, and the result of comparison (E (x, y)) is greater than the decision value. If small, the result value E (x, y) is recorded as a new determination value and the displacement (dx, dy) shifted for the motion vector search provided from the controller is recorded as a new motion vector. And a comparison determination unit for outputting and resetting the currently recorded motion vector (M _X , M _Y ) when a control signal (C2) is input, wherein x, y, i, and j are arbitrary amounts including 0. Motion using similarity of gradient and intensity of an image characterized in that it is an integer Vector navigation device. The intensity matching equation used in the intensity matching unit is And Bx and By are arbitrary positive integers, each of which is a horizontal and vertical size of the average value region B, wherein the motion vector search apparatus employs similarity of gradient and intensity of an image. The method of claim 1, wherein the gradient matching equation used in the gradient matching unit is And Bx and By are arbitrary positive integers, each of which is a horizontal and vertical size of the average value region B, wherein the motion vector search apparatus employs similarity of gradient and intensity of an image. According to claim 1, wherein the determination formula used in the comparison determination unit is And W ₁ and W ₂ are weights. The motion vector search apparatus using the similarity of the gradient and intensity of an image.