KR960012935B1 - Scene change detecting method and apparatus utilizing mean-absolute difference value between image frames - Google Patents

Abstract

The detector comprises the first absolute calculating means(32), an accumulating means(33), a production means(34), the first latch means(36), the second absolute difference means(35), the first comparing means(37), the second comparing means(38), a selecting means(39), and the second latch means(40). The method includes the steps of; investigating scene change variable, determinating scene change using average absolute difference, determinating scene change using difference of MAD and ending.

Description

Scene change detector using difference of average absolute difference between image frames and its detection method

1 is a conventional scene change detection flowchart.

2 is an exemplary diagram showing the difference between the mean absolute difference and the mean absolute difference in successive video signals in order to clarify the present invention.

3 is a block diagram of a scene change detector according to the present invention.

4 is a process flow diagram for a scene change detection method according to the present invention.

5 is a process flow diagram for scene change determination using the mean absolute difference according to the present invention.

6 is a process flow diagram for scene change determination using the difference of the mean absolute differences according to the present invention.

* Explanation of symbols for main parts of the drawings

31: frame memory 32, 35: absolute difference calculator

33: accumulator 34: multiplier

36,40: latch 37,38: comparator

39: selector

The present invention relates to a scene change detector using a difference of MAD (dMAD) of a mean absolute difference (MAD) between image frames, and a detection method thereof.

In general, scene change information is useful for predictive encoding of a digital video signal. When a scene change occurs, prediction of a current frame for a previous frame becomes meaningless, and therefore, intraframe coding must be performed after the scene change.

In particular, the scene change information is a macro block unit, and a bit rate control of an image encoder that adaptively applies an intra-frame encoding mode or an interframe predictive coding mode according to its characteristics. It is most useful in the process. If there is no scene change detection process, in the case of the first frame after the scene change, most macro blocks are actually processed in the intra frame encoding mode even in the inter frame prediction encoding mode. In this case, an amount of data far exceeding the target number of bits is generated, resulting in a problem in that the bit rate is unbalanced and the image quality after the current frame is excessively degraded. Therefore, in adaptive predictive encoding, it is very important to make a scene change determination before encoding each frame to determine whether the frame is encoded in the frame or the inter frame. For example, the differences between frames, interframe correlations, average brightness of frames, and quantization curves are often used.

1 is a flowchart of the prior art in which a scene change determination is made using the difference between frames.

The average absolute difference between the previous frame and the current frame is calculated (1) and if the value is greater than the threshold (2), the scene is switched (3). If it is not greater than the threshold, there is no scene transition (4). The scene change information obtained in the process (3, 4) can be used for bit rate adjustment of the encoder. When the scene change determination is made using the conventional method, there is a high possibility that a scene having a large movement is mistaken as almost every frame scene change even though there is no scene change. This is because when the motion is large, the MAD value of each frame becomes larger than a given threshold, making it difficult to distinguish from the actual scene change.

Disclosure of Invention The present invention devised to solve the above problems is to provide a simple image transition detector and a detection method having a low probability of determination error by using a difference value of the average absolute difference between image frames.

Device forming the present invention, in order to achieve the above object, the absolute difference between the input signal (f _a) a frame memory, the input signal (f _a) and the output signal (f _a-1) of the frame memory for storing A first absolute difference calculating means for outputting, an accumulation means for accumulating the absolute difference, a multiplication means for obtaining an average absolute difference by multiplying a constant factor (FACTOR) as an input, and an average absolute difference through the multiplication means The difference dMAD (n) of the average absolute difference between the first latch means for storing for one frame time, the output MAD (n) of the multiplication means and the output MAD (n-1) of the first latch means. A second absolute means for calculating)), a first comparison means for determining a scene change using an absolute value from the second absolute means and the threshold Th2, and an average absolute difference of the corresponding frame from the multiplication means A second transition scene decision made by comparing the bay with the threshold Th1 Selection means for determining which of the comparison means, the first and second comparison means to use, and outputting the scene change information; and providing the selection means with the scene change information of the previous frame stored therein. And second latch means for determining the comparison means.

In addition, the method applied to the above configuration may calculate the average absolute difference of the input nth image frame and determine the scene change to be used in the current frame according to the scene change state (scene_change (n-1)) of the immediately preceding frame. After performing the first step of checking whether the variable is '1' and after performing the first step, if the scene change variable value is '1', the current frame corresponds to the second frame of the new scene, and thus the scene change determination is performed using only the average absolute difference. After performing the second step, the third step of performing a scene change determination using the difference of the average absolute difference as the scene change variable, if the scene change variable value is not '1', and the second step and the second step After performing the third step, the scene change state and the average absolute difference of the n th frame are recorded and used to process the n + 1 th frame, and then the fourth step ends.

The average absolute difference between the brightness signal (Y) components of two frames is defined as follows.

f _a (i, j): video signal of the nth frame M: number of pixels per line; N: lines per frame

Here, the factor FACTOR is a constant 1 / (M × N) for obtaining an average. MAD corresponds to the first derivative (df / dt, where f is a video signal in units of frames), and when there is a scene change between frames, it has a large slope at that portion. Differential derivatives, that is, second derivatives (d ² f / dt ² ) with respect to the original signal f, have a peak value once the slope of the MAD is large, and thus it is easy to detect a change part using a threshold value. Difference in average absolute difference between frames, which is the difference of MAD between frames, Is the second derivative. Therefore, it is very useful to use dMAD as a transition detection variable in successive frames. In the case of a large-motion scene, each MAD value is large and the dMAD value is small. In other words, the use of dMAD can reduce the scene change determination error in the case of a large scene.

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

2 shows MAD and dMAD in successive video signals for explaining the present invention, in which f _n represents the video signal of the nth frame.

As shown in the figure, since the second frame of the new scene has no meaning, dMAD is used only in this case MAD of the current frame. For example, if there is a scene change in the nth frame, MAD (n) is used to calculate dMAD (n + 1) of the n + 1th frame. However, since MAD (n) is an average absolute difference between f _n-1 which is the last frame before the new scene and f _{n which} is the first frame of the new scene, MAD (n) has a large value regardless of the transition of f _{n + 1} . If there is no transition at f _{n + 1} , MAD (n +1) has a small value, so dMAD (n +1) has a large value, which is incorrect because there is a transition at f _{n + 1} . . Therefore, dMAD cannot be used to determine the transition in the second frame of the new scene after the transition.

3 is a block diagram of a scene change detector according to the present invention, in which 31 is a frame memory, 32 and 35 are absolute difference calculators, 33 is an accumulator, 34 is a multiplier, 36 and 40 is a latch, and 37 and 38 are Comparator, 39 represents a selector, respectively.

The inventors scene change detector as shown in the figure, an input signal (f _n) for storing the frame memory 31, the input signal (f _n) and the output signal (f _n-1 of the frame memory 31 to Absolute difference calculator 32 that outputs the absolute difference of), the accumulator 33 accumulating the absolute difference, and the output of the accumulator 33 are input and multiply by a constant factor (FACTOR) to obtain the average absolute difference MAD. Multiplier 34, where the factor is generally the total pixel size of one frame to obtain an average value, but in the present invention, a shift register is used to simplify the hardware implementation. Differences from the values used were taken into account in the thresholds (Th1, Th2). A latch 36 for storing the average absolute difference through the multiplier 34 for one frame time, an output MAD (n) of the multiplier 34 and an output MAD (n-1) of the latch 36 The absolute difference calculator 35 which calculates the difference value (dMAD (n)) of the average absolute difference with (), and the comparator which determines a scene transition using the threshold value Th2 of the absolute difference from the absolute difference calculator 35 ( 37) a scene is determined by determining which of the comparator 38 and the comparators 37 and 38 to perform a scene change determination by comparing only MAD (n) of the corresponding frame from the multiplier 34 with a threshold Th1. A latch 40 for determining the comparators 37 and 38 by providing a selector 39 for outputting switching information and scene change information (scene_change (n-1)) of a previous frame stored by the selector 39. (For example, the selector 39 adopts the comparator 37 because the scene change information n-1 has no scene change in the '0' dlaus n-1th frame. It consists of more than dMAD and to process the n-th frame using a (n)).

4 is a flowchart illustrating a scene change detection method according to the present invention.

The MAD (n) of the input nth image frame f _n is calculated (41) to obtain a value that varies the scene change determination variable to be used in the current frame according to the scene change information (scene_change (n-1)) of the immediately preceding frame ( 42).

If the scene change variable value is '1' in the process 42, since there was a scene change in the previous frame f _n-1 , the current frame corresponds to the second frame of the new scene and the scene change decision is made using only MAD (n). (43) If the transition variable value is not '1', use dMAD (n) as the transition variable (44) and record the transition state scene_change (n) and MAD (n) of the nth frame as a result, and then n + It is used for the processing of the first frame and ends (45).

5 is a flowchart illustrating a scene change determination method using an average absolute difference (MAD) according to the present invention.

If the MAD (n) of the current frame is compared with the threshold (51) and lower than the threshold (52), the change between frames is small, so it is determined that there is no transition and the transition state (scene_change (n)) is set to '0'. do. If MAD (n) is greater than or equal to the threshold (53), since the change between frames is large, it is determined that the scene has changed, and the transition state is set to '1'.

6 is a flowchart illustrating a scene change determination method using a difference value of an average absolute difference according to the present invention.

The difference dMAD (n) of the mean absolute difference is obtained from the current mean absolute difference MAD (n) and the previous mean absolute difference MAD (n-1) (61). If it is less than the threshold value of the investigation 62, it is determined as a continuous scene without scene change, and the scene change state scene_change (n) is set to '0' (63). If dMAD (n) is greater than or equal to the threshold, it is determined that the scene has been switched, and the scene change state is set to '1', and the process ends (64).

As described above, the present invention has the effect of greatly reducing the cost of hardware implementation while achieving almost the same performance as the motion estimation technique as the scene change detection method.

Claims (4)

Input signal frame to store (f _n) merori 31, wherein the input signal (f _n) of the first absolute difference calculating means for outputting the absolute difference between the output signal (f _n-1) of the frame memory 31 (32), multiplication means (33) for accumulating the absolute difference, multiplication means (34) for obtaining an average absolute difference by multiplying a constant factor (FACTOR) as an input and outputting the accumulation means (33), and multiplication means (34) The first latch means 36 for storing the average absolute difference through) for one frame time, the output MAD (n) of the multiplication means 34 and the output MAD (n of the first latch means 36. -2) using the second absolute value calculating means 35 for obtaining the difference value dMAD (n) of the average absolute difference with the absolute value and the threshold value Th2 from the second absolute value calculating means 35 First comparison means 37 for determining a scene change, second comparison means 38 for making a scene change determination by comparing only the average absolute difference of the corresponding frame from the multiplication means 34 with a threshold Th1, Selection means 39 for outputting the scene change information by determining which of the first and second comparison means 37 and 38 to use, and the scene change information of the previous frame stored in the selection means 39. And second latch means (40) for determining the first and second comparison means (37, 38) to provide the first and second comparison means (37, 38). Frame memory 31, first absolute calculation means 32, accumulation means 33, multiplication means 34, first latch means 36, second absolute calculation means 35, first comparison means (37), a scene change detection method applied to a scene change detector using a difference value of the average absolute difference between image frames including the second comparison means 38, the selection means 39, and the second latch means 40. And calculating a mean absolute difference of the n-th input image frame and checking whether the scene change determination variable to be used in the current frame is '1' according to the scene change state (scene_change (n-1)) of the immediately preceding frame. After performing the steps 41 and 42 and the first steps 41 and 42, when the scene change variable value is '1', the scene change determination is performed using only the average absolute difference since it corresponds to the second frame of the new scene of the current frame. After performing the second step 43 and the first steps 41 and 42, if the scene change variable value is not '1', the difference of the average absolute difference is obtained. After performing the third step 44 which determines the transition using the scene change variable, and after performing the second and third steps, the transition state and the average absolute difference of the n th frame are recorded and the n + 1 th frame is recorded. And a fourth step (45) to be used in the processing and to end the scene change detection method using the difference value of the average absolute difference between video frames. 3. The method of claim 2, wherein said second step (43) comprises: After performing the first process 51 comparing the absolute difference of the current frame with the threshold value and performing the first process 51, if it is lower than the threshold value, it is determined that there is no scene change and the scene change state scene_change (n) is set to '0. And a second process 52, 53 for determining that the scene is changed and setting the scene change state to '1' if it is greater than or equal to the threshold value, and using the difference value of the average absolute difference between video frames. Way. The method of claim 2, wherein the third step (44) comprises: After performing the first steps (61, 62) and the first steps (61, 62) of obtaining the difference value of the mean absolute difference from the current mean absolute difference and the previous mean absolute difference and checking whether the difference of the mean absolute difference is smaller than the threshold value, If it is less than the threshold, it is determined as a continuous scene without a transition, and the transition state (scene_change (n)) is set to '0', and if it is greater than or equal to the threshold, it is determined that the scene is changed and the transition state is set to '1'. And a second process (63, 64) of terminating the scene change detection method using the difference value of the average absolute difference between video frames.