JPH04288789A - Motion compensation predicting system - Google Patents

Abstract

PURPOSE:To obtain the motion compensation predicting system to miniaturize a device without reducing accuracy for predicting motion compensation so much even when a motion vector is detected while dividing the detection into N (N>=2) stages and an evaluation function calculating amount and the number of times for comparison are reduced in the case of predicting the motion compensation. CONSTITUTION:A reference block is detected while dividing the detection into the plural N stages and in the case of detection on an L-th stage, the value can be increased/decreased according to the amount of information as needed. A threshold value is provided an an adaptive motion compensation predicting circuit 11 is provided according to the threshold value so that when the evaluation of an evaluation function value is higher than the threshold value, the relevant motion vector can be defined as an output to the next stage and that when the evaluation is low, the arbitrary number of motion vectors based on the evaluation function value can be outputted to the next stage or evaluation can be stopped.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement in a motion compensation prediction method often used in image processing apparatuses, particularly in image coding and transmission apparatuses for image communication such as video conferences and video telephones.

0002

[Prior Art] In general, image transmission used for image communication such as video conferencing or video telephone calls involves a huge amount of image information, but due to the line speed and transmission cost during transmission, it is difficult to transmit image information. Compression encoding devices that reduce (compress) the amount of data have been put into practical use. Among these, the motion compensation method is known as a method for compressing the amount of information as having a high degree of compression.

[0003] A conventional image transmission device will be explained below with reference to the drawings. FIG. 5 shows a basic block diagram of a conventional image transmission apparatus to which a motion compensated predictive interframe quantization method is applied. In FIG. 5, the conventional image transmission device is roughly divided into a preprocessing circuit 1, a scan conversion circuit 2, a motion compensation prediction circuit 3, a quantization circuit 5, and a variable length encoding circuit 6.
, an inverse quantization circuit 8 , and a frame memory 10 .

Next, the operation and signal flow will be explained. The preprocessing circuit 1 performs A/D conversion on an image input signal 100 and then performs image preprocessing to generate a time division multiplexed (TDM) signal 101.
Convert to The scan conversion circuit 2 has a TD as shown in FIG.
A k-dimensional block signal 102 is created by dividing adjacent pixels of the M signal 101 into k blocks. The motion compensation prediction circuit 3 creates a plurality of blocks as reference blocks including a block signal 102 and a block corresponding to the same position as the block signal 102 stored in the frame memory 10, and generates a reference block signal 103b and its shift. A reference block generation circuit 3a that generates a vector (displacement information from the same position) 103a, and a block signal 1
02 and the reference block signal 103b, a distortion calculation (for example, Euclidean distortion, absolute value distortion, etc.) is performed, a block with the minimum distortion value is searched (detected) from among the reference blocks, and the block of the reference block is The shift vector is output as a motion vector 104a. Further, a motion vector detection circuit 3b outputs a predicted signal 104b from the frame memory 10 based on the motion vector 104a.
It is made up of.

The motion vector 104a described above is provided to the variable length encoding circuit 6 and the frame memory 10. The predicted signal 104b, which is the output of the frame memory 10, is output to the subtracter 4 and the adder 9, and the difference block signal 105 obtained by the subtracter 4 is applied to the quantization circuit 5. The quantization circuit 5 maps the differential block signal 105 to a quantization table to obtain a quantization number 106. The variable length encoding circuit 6 inputs the motion vector 104a and the quantization number 106, converts and multiplexes them into variable length codes assigned to each, and outputs a code multiplexed signal 107.

[0006] The transmission buffer memory 7 temporarily stores the code multiplexed signal 107 and outputs it to the transmission path as a code output 108 in synchronization with the transmission speed. On the other hand, quantization number 10
6 is also input to the inverse quantization circuit 8, where it is quantized and inversely transformed to reproduce the differential block signal 109. The adder 9 receives the predicted signal 104b and the reproduced difference block signal 1.
09 is added, and the block signal 110 is reproduced. The frame memory 10 stores the reproduced block signal 110 and outputs it to the motion compensation prediction circuit 3.

By the way, the operation of the motion compensation prediction circuit 3 described above will be described in some detail. The above circuit selects an optimal shift vector as a motion vector from among the created reference blocks. At this time, if the input image is an image with little movement, such as an image signal of a video conference, there is a high probability that the vicinity of the reference block at the same position will be selected. However, if the movement becomes faster, you will have to select from a wider range of blocks than just the original blocks. As an optimal method for detecting a motion vector in such a case, for example, there is a method in which a plurality of shift vectors are set in advance around the same position and a distortion calculation is performed on all the shift vectors to obtain a motion vector.

[0008] Furthermore, in order to reduce the number of calculations with many shift vectors and to reduce the scale, a method has been proposed in which detection is performed in multiple stages from coarse to fine. For example, as shown in Japanese Patent Publication No. 55-158784 (interframe coding device), "motion vector detection is divided into N stages (N≧2) and the Lth stage (L=
1, 2, ..., N-1), the (L+1)th shift vector group is determined based on the evaluation function obtained for the determined Lth shift vector group, and in the Nth step, the determined One example is a method in which one shift vector is detected as a motion vector based on the evaluation function determined for the Nth shift vector group.

Further, when absolute value distortion is employed as the criterion d for determining the shift vector in the motion vector detection circuit 3b, the one that minimizes the following equation is selected.

0010

[Math 1]

[0011]

[Problems to be Solved by the Invention] Since the conventional image transmission device using motion compensation prediction has the above configuration, the amount of evaluation function calculations and the number of comparisons required for motion vector detection in motion compensation prediction are reduced. The larger the number, the larger the device becomes. In addition, the above-mentioned published patent publication No. 55-158784 (
In the method of reducing the amount of calculation by performing motion vector detection in N stages as shown in the Interframe Coding Device), the shift vector of the next stage is always selected at each stage, so the evaluation function Although the motion vector corresponding to the maximum (minimum) point of is detected, there is no guarantee that the optimal motion vector will be detected, and the problem is that motion compensation prediction accuracy decreases due to misdetection of the shift vector in the Lth stage (especially in the stage close to the first stage). There is a point.

0012

[Means for Solving the Problems] In order to solve the above problems, the present invention provides the following steps when performing vector detection in N stages:
When detecting the L-th stage, a threshold value that increases or decreases based on the information amount signal is set, and when only the best evaluation function value among the shift vector group exceeds the set threshold value, only that value is set as the L-stage shift vector; An adaptive motion compensation prediction circuit is provided which selects a plurality of shift vectors with small evaluation function values or stops evaluation at the L-th stage when there are no or a plurality of shift vectors exceeding the threshold value.

[0013]

[Operation] The motion compensation prediction method according to the present invention detects shift vectors in N stages, provides a threshold value, makes it variable up and down, and has an adaptive motion compensation prediction circuit that selects the number of shift vectors. , the next-stage shift vector is selected to be the optimal one or a plurality of possible ones.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A preferred embodiment of an image transmission apparatus to which the motion compensation prediction method of the present invention is applied will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of an image transmission apparatus using a motion compensation prediction method according to the present invention. FIG. 2 is an explanatory diagram of motion vector detection according to a first control example using the motion compensation prediction method according to the present invention. 3 and 4 are explanatory diagrams of motion vector detection according to a second control example of the motion compensation prediction method according to the present invention. In the drawings, parts that are the same as those of the conventional device are designated by the same reference numerals, and their explanations will be omitted.

The novel part of this embodiment is the provision of an adaptive motion compensation prediction circuit 11. The adaptive motion compensation prediction circuit 11 includes a reference block generation circuit 11a, an evaluation function calculation circuit 11b, and a motion vector detection control circuit 11c. The reference block generation circuit 11a reads the image data 113 from the frame memory 12 and generates a reference block of the shift vector group 112 specified by the motion vector detection control circuit 11c. The evaluation function calculation circuit 11b calculates an evaluation function, for example, a distortion value, of the block signal 102 and the reference block group 114 from the reference block generation circuit 11a, and calculates a distortion value 115 that is the evaluation value of the reference block. is output to the motion vector detection control circuit 11c.

The motion vector detection control circuit 11c includes:
Control for detecting motion vectors in N stages (N≧2) in the adaptive motion compensation prediction circuit 11 is performed by comparing the threshold value TL with the evaluation function value 115, and the finally obtained motion vector 116 is is output to the variable length encoding circuit 6 and frame memory 12. At this time, the amount of information generated 124 counted in the transmission buffer memory 7
The threshold value TL is raised or lowered to adjust the calculation amount and prediction accuracy of motion compensation prediction. A predicted block signal 117 is output from the frame memory 12 based on the motion vector 116.

Next, two preferred control examples for the overall operation of the adaptive motion compensation prediction circuit 11 will be explained with reference to FIGS. 2, 3, and 4. To simplify the explanation, N=3 and the evaluation function is value (the smaller the value, the higher the evaluation). In the first control example, when performing motion vector detection in three stages, the smallest distortion value EL in the L-stage shift vector group is compared with the L-stage threshold value TL, and EL≦ When the condition of TL (or EL < TL) is satisfied, only the shift vector that gives EL is used as the L-stage shift vector, and when the condition is not satisfied, an arbitrary number (for example, 2) with the smallest distortion value is selected and used as the L-stage shift vector. Shift vector(s). Then, from these, the (L+1)th stage shift vector group is determined, and the third
One shift vector is used as a motion vector in each stage.

That is, as an example, FIG. 2(a) shows the first stage (L=1), (b) shows the second stage (L=2), and FIG. 2(c) shows the third stage (N=3: final stage). FIG. First, in the first stage, calculations are performed on five shift vectors A0, B0, C0, D0, and E0 (E0 is the same position with no movement), and shift vector D0 indicating the minimum distortion value E1 is selected. It is assumed that the relationship with the first stage threshold value T1 is E1≦T1. In this case, the condition is met, so the first stage shift vector is only D0. In the second stage, D1, D2, D3, D4 are newly set based on the first stage shift vector D0, and calculations are performed on five shift vectors including D0, and the shift vector D1 indicating the minimum distortion value E2 is calculated.
is selected. At this time, it is assumed that the relationship with the second stage threshold value T2 is E2>T2. In this case, the condition does not hold, so two second-stage shift vectors, D1 and D4, are selected from those with the smallest distortion values. Then, in the final stage, the second
D11, D12, based on stage shift vectors D1, D4,
D13, D14, D41, D42, D43, and D44 are newly set, and calculations are performed on 10 shift vectors including D1 and D4. Shift vector D41 indicating the minimum distortion value E3 is detected as a motion vector, and selection is completed. .

In the second control example, when performing motion vector detection in three stages, each evaluation function value ELi (i: L-stage shift vector identifier) of the L-stage shift vector group and the L-stage shift vector All the shift vectors that satisfy the condition of ELi≦TL (or ELi<TL) are compared with the threshold value TL, and the (L+1)th stage shift vector group is determined as the Lth stage shift vector, and the shift that satisfies the condition is determined. If no vector exists, the shift vector that gives the smallest evaluation function value is used as the motion vector, and the search is stopped at the Lth stage.

The example (part 1) of FIG. 3 is a case where there is at least one shift vector that satisfies the condition in the L-th stage. FIG. 3(a) is a diagram explaining a specific example of the first stage (L=1), (b) is the second stage (L=2), and (c) is the third stage (N=3: final stage). It is. First, in the first stage, A0, B0
, C0, D0, E0 (E0 is the same position with no movement), and the relationship with the first stage threshold T1 is ED0≦T1<EE0<EC0<EA
Suppose that 0<EB0. At this time, the only first stage shift vector that satisfies the condition is D0. In the second stage, D1, D2, D3,
Assume that D4 is newly set, calculations are performed on five shift vectors including D0, and the relationship with the second stage threshold T2 becomes ED0<ED1<ED4≦T2<ED2<ED3. In this case, the second stage shift vectors for which the condition is satisfied are D0, D1, and D4. Subsequently, in the final stage, D01 is set based on the second stage shift vectors D0, D1, D4.
(=D13), D02, D03, D04 (=D42),
D11, D12, D14, D41, D43, and D44 are newly set (select one of the overlapping shift vectors), calculations are performed on 13 shift vectors including D0, D1, and D4, and the minimum distortion value E3 is calculated. The indicated shift vector D41 is detected as a motion vector, and the selection is completed.

The example (part 2) in FIG. 4 is a case where there is no shift vector that satisfies the condition in the second stage. FIG. 4(a) shows the first stage (L=1), and FIG. 4(b) shows the second stage (L=1).
FIG. 4 is a diagram illustrating a specific example of L=2). First, in the first stage, 5 shift vectors A0, B0, C0, D0, E0 (
E0 is the same position with no movement).
The relationship with the first stage threshold T1 is ED0≦T1<EE0
Assume that <EC0<EA0<EB0. At this time, the only first stage shift vector that satisfies the condition is D0. In the second stage, D1,
D2, D3, and D4 are newly set, and calculations are performed on 5 shift vectors including D0, and the relationship with the second stage threshold T2 is T2<ED0<ED1<ED4<ED2<ED3
If so, then the condition will not hold. Therefore, the selection ends at the second stage, and the shift vector D0 indicating the minimum distortion value E3 is detected as the motion vector.

In the above two embodiments, the first stage threshold value
If the condition is not satisfied by comparison with TL, high-speed processing may be performed in which E0, which is at the same position (no movement), is selected as the motion vector and the search is stopped.

Although the motion compensation block size (H pixels×V lines) is not specified in the above embodiment, it can be set to any block size. generally 16
Pixels×16 lines are often used. Also, the number of detection stages N
, shift vector group, etc. can have similar effects even if other settings are used. Further, the quantization circuit and the inverse quantization circuit can produce the same effect even if the circuit employs other methods such as discrete cosine transform+quantization or vector quantization. Further, the same effect can be obtained even if the L-stage threshold value TL is made common to each stage. Further, in the second control example, all shift vectors whose distortion values are smaller than the L-stage threshold TL are used as L-stage shift vectors, but the same effect can be achieved even if any number of shift vectors are selected. Moreover, the same effect can be obtained even if the threshold value T0 for the same position is made common to other threshold values TL, especially the first stage threshold value T1. In addition, motion vectors are detected while controlling the L-stage threshold TL depending on the amount of information generated, but other parameters that affect the encoding situation, such as transmission speed and number of frames transmitted per unit time, are also considered. A similar effect can be obtained by controlling the number of frames depending on the situation of falling pieces or the like.

[0024]

As described above, according to the present invention, shift vector detection is performed in N stages, detection in the L stage is performed by setting a threshold value that increases or decreases depending on the amount of information, and furthermore, the evaluation function value As a result of the comparison, the output is determined to be only one, multiple, or the evaluation is discontinued, so it is possible to eliminate false detection and determine the necessary shift vector while keeping the amount and number of calculations of the evaluation function small, and in some cases, the detection time can be It has the effect of further shortening the time.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of an image transmission device using a motion compensation prediction method according to the present invention.

FIG. 2 is an explanatory diagram of motion vector detection according to a first control example of the compensation prediction method according to the present invention.

FIG. 3 is an explanatory diagram (Part 1) of motion vector detection according to a second control example of the compensated prediction method according to the present invention.

FIG. 4 is an explanatory diagram (part 2) of motion vector detection according to a second control example of the compensation prediction method according to the present invention.

FIG. 5 is a basic block diagram of a conventional image transmission device to which a motion compensated predictive interframe quantization method is applied.

FIG. 6 is an explanatory diagram of blocking of a scan conversion circuit.

[Explanation of symbols]

1 Pre-processing circuit 2 Scan conversion circuit 4 Subtraction circuit 5 Quantization circuit 6 Variable length encoding circuit 7 Transmission buffer memory 8 Inverse quantization circuit 9 Adder circuit 11 Adaptive motion compensation prediction circuit 12 Frame memory

Claims (1)

[Claims] 1. An image input signal is divided into blocks to obtain a block signal, shifted as necessary to form a plurality of reference block groups, and an approximate reference block is obtained by calculating an evaluation function between the block signal and the reference block signal. In motion compensated prediction that detects and outputs a motion vector, the above-mentioned detection is performed in a plurality of N stages, and at the time of the L-th stage detection, a threshold value is provided whose value can be changed up or down depending on the amount of information as necessary, If the evaluation function value is higher than the threshold value, output the motion vector giving the above evaluation to the next stage, and if the evaluation is lower, output an arbitrary number of motion vectors to the next stage based on the evaluation function value. , or a motion compensation prediction method characterized by being provided with an adaptive motion compensation prediction circuit that sets a threshold, or stops the evaluation and leaves the pre-evaluation vector as it is, and outputs the only motion vector at the final stage.