JP3804764B2 - Motion compensated prediction singular value expansion coding apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a motion compensated prediction singular value expansion encoding apparatus, and more particularly to a motion compensated prediction singular value expansion encoding apparatus capable of encoding a moving image with high efficiency.
[0002]
[Prior art]
Conventionally, a hybrid encoding method combining motion compensation prediction and DCT is often used as a moving image encoding apparatus. In FIG. An example of an encoding apparatus standardized by H.263, etc. is shown.
[0003]
An image input from a television camera or the like is subjected to motion detection in units of blocks with the image of the previous frame stored in the frame memory 39 in the motion detection unit 31. As a motion detection method, a method in which a displacement that minimizes the sum of squares of errors between frames is used as a motion vector, a method in which a pixel value is obtained from a time-space gradient, a method in which an FFT is obtained, and the like are known. The motion compensation prediction unit 40 uses the motion vector obtained by the motion detection unit 31 to predict the prediction image corresponding to the input image from the image of the previous frame in units of blocks.
[0004]
The subtraction unit 32 receives the input image and the prediction image output from the motion compensation prediction unit 40, and the subtraction unit 32 subtracts the prediction image from the input image and outputs a prediction error. The DCT unit 33 performs DCT on the input prediction error to obtain a DCT coefficient. The quantization unit 34 performs quantization on the DCT coefficient, and the entropy coding unit 35 performs entropy coding on the quantized DCT coefficient. The entropy encoded data is stored in a transmission or storage device.
[0005]
The quantized DCT coefficient is input to the inverse quantization unit 36 and inversely quantized, and then the inverse DCT unit 37 performs inverse DCT. The inverse DCT signal is decoded by being added by the adder 38 with the predicted image from the motion compensation prediction unit 40 and stored in the frame memory 39.
[0006]
[Problems to be solved by the invention]
However, although the above-described encoding apparatus can perform efficient encoding, further efficiency improvement is demanded from the standpoint of improving image quality, particularly at the time of low bit rate encoding.
[0007]
The present invention has been made in view of the above-described prior art, and an object thereof is to provide an adaptive orthogonal transform coding apparatus that can perform coding with higher efficiency than a conventional coding apparatus.
[0008]
[Means for Solving the Problems]
In order to achieve the above-described object, the present invention performs SVD conversion using the SVD obtained from the candidate prediction image block, as a prediction error between the block of the current image and a plurality of predetermined candidate prediction image blocks of the previous image. An SVD motion detection unit that obtains an SVD conversion base and a motion vector MV for the candidate prediction image block having the smallest code amount, and obtains a code amount by quantizing and entropy encoding the result of the SVD conversion; An SVD orthogonal transform unit that orthogonally transforms a prediction error, which is a difference from a predicted image that has been motion compensated with the motion vector MV, using the SVD transform base obtained by the SVD motion detection unit, and the SVD orthogonal transform unit. It is characterized in that it comprises means for quantizing and entropy encoding the conversion result.
[0009]
According to this feature, SVD orthogonal transform with higher coding efficiency can be performed instead of the conventional DCT, and the SVD transform base used for the SVD orthogonal transform does not need to be sent to the receiving side. Encoding can be performed with higher efficiency than the encoding device.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.
[0011]
In the figure, an input image is input to the SVD motion detection unit 1 in block units (hereinafter referred to as an input image block A), and a predicted image one frame before from the frame memory 10 is referred to as a block unit (hereinafter referred to as a predicted image). (Referred to as block B). Although the details of the function of the SVD motion detection unit 1 will be described later, the SVD motion detection unit 1 performs motion detection using SVD (singular value expansion) between the input image block A and the predicted image block B. The SVD motion detection unit 1 outputs a motion vector MV and an SVD conversion base for the motion vector MV. The motion vector MV is sent to the motion compensation prediction unit 2 and the receiving side, while the SVD transform base is sent to the SVD orthogonal transform unit 4 and the inverse SVD orthogonal transform unit 8.
[0012]
The subtraction unit 3 receives the input image block A and the prediction image block C corresponding to the motion vector MV output from the motion compensated prediction unit 2, and the subtraction unit 32 predicts from the input image block A. The prediction error is output by subtracting the image block C. The SVD orthogonal transform unit 4 subjects the input prediction error to orthogonal transform using the SVD transform base sent from the SVD motion detection unit 1 to obtain an SVD orthogonal transform coefficient. The quantization unit 5 performs quantization on the SVD orthogonal transform coefficient, and the entropy coding unit 6 performs entropy coding on the quantized coefficient. The entropy encoded data is transmitted along with the motion vector MV or stored in a storage device.
[0013]
The quantized SVD orthogonal transform coefficient is input to the inverse quantization unit 7 and inversely quantized, and then subjected to inverse SVD by the inverse SVD orthogonal transform unit 8. The inverse SVD signal is decoded by adding the predicted image from the motion compensation prediction unit 2 to the addition unit 9 and stored in the frame memory 10.
[0014]
Next, the function of the SVD motion detector 1 will be described with reference to the flowchart of FIG.
[0015]
In step S1, a certain number i = 1 is set, and in step S2, a candidate prediction image block Bi for the input image block A is obtained, and SVD conversion bases [S _v ] and [S _H ] for the Bi are obtained. The SVD conversion base [S _v ] and [S _H ] for the Bi can be obtained as follows.
[0016]
If the candidate predicted image block B is a pixel value matrix [Bi] composed of N × N pixels, the covariance matrix [C _H ] in the horizontal direction is [C _H ] = [Bi] [Bi] ^T. . The covariance matrix [C _v ] in the vertical direction is [C _v ] = [ ^B i] ^T [ ^B i]. Here, [Bi] ^T represents a transposed matrix.
[0017]
Next, a corresponding orthogonal transform (KLT) basis ([S _H ], [S _V ]) is obtained from the covariance matrices [C _H ], [C _v ]. Then, the pixel value matrix [Bi] is SVD expanded by the following equation.
_{[BiSVD] = [S V]} [Bi] [S H]
[0018]
At this time, if the matrix [Bi] is an N × N matrix, the SVD expansion coefficient [BiSVD] is also an N × N matrix, and all components other than the diagonal components are zero. This conversion is called SVD, and is an optimal conversion when a rectangular block is expressed by two independent horizontal / vertical orthogonal transforms. That is, the pixel matrix [Bi] in the rectangular block can be expressed with the smallest number of coefficients. Therefore, efficient encoding can be performed.
[0019]
Next, in step S3, a difference between the candidate predicted image block Bi and the input pixel block A, that is, a prediction error Ci is obtained. Let Ci = Bi-A. In step S4,
The vertical and horizontal SVD conversion bases [S _V ] and [S _H ] obtained in step S2 are applied to the prediction error Ci. When the result is [CiSVD], [CiSVD] = [S _V ] [ _C i] [S _H ].
[0020]
In step S5, the SVD expansion coefficient [CiSVD] of [Ci] is quantized and entropy encoded. In step S6, the quantization and entropy-encoded SN and code amount Ei are obtained. In step S7, it is determined whether i = Z is established. Here, Z is the total number of candidate predicted image blocks Bi for block A. When this determination is negative, the process proceeds to step S8, and 1 is added to i. Then, returning to the step S2, the operations of the steps S2 to S7 are repeated again.
[0021]
As a result, if the determination in step S7 becomes affirmative, the process proceeds to step S9 to obtain a candidate predicted image Bimin having the smallest code amount Ei. In step S10, as shown in FIG. 3, the displacement of the block A with respect to the predicted image Bimin is set as a motion vector MV. The motion vector MV is sent to the motion compensation prediction unit 2 and the receiving side. In step S11, the SVD orthogonal transform base [S _V ] and [S _H ] for the predicted image Bimin are sent to the SVD orthogonal transform unit 4 and the inverse SVD orthogonal transform unit 8.
[0022]
As described above, according to the encoding apparatus of the present embodiment, orthogonal transform is performed on each block using SVD having higher conversion efficiency than DCT. When SVD is applied, it is generally necessary to encode and send the SVD transform base calculated for each block. However, in the present embodiment, instead of the SVD transform base for the prediction error, prediction is not performed. Since the SVD orthogonal transform base [S _V ] and [S _H ] for the image Bimin are used, it is not necessary to send them. That is, since the decoding device on the receiving side for the encoding device receives the motion vector MV for the predicted image Bimin obtained in step S10, the predicted image Bimin can be reproduced from the motion vector MV. Therefore, the SVD orthogonal transform bases [S _V ] and [S _H ] can be reproduced from the predicted image Bimin, and it is not necessary to send the SVD orthogonal transform bases [S _V ] and [S _H ] to the receiving side.
[0023]
【The invention's effect】
As is apparent from the above description, according to the present invention, each block of the current image can be subjected to orthogonal transform using SVD having higher conversion efficiency than DCT, and the SVD orthogonal transform can be performed. It is not necessary to send the SVD conversion base used in the above to the receiving side or the recording medium, so that highly efficient encoding can be performed.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.
FIG. 2 is a flowchart showing an operation of the SVD motion detection unit in FIG. 1;
FIG. 3 is an explanatory diagram of a motion vector MV obtained by an SVD motion detection unit.
FIG. 4 is a block diagram showing a configuration of a conventional apparatus.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... SVD motion detection part, 2 ... Motion compensation prediction part, 3 ... Subtractor, 4 ... SVD orthogonal transformation part, 5 ... Quantization part, 6 ... Entropy encoding part , 7... Inverse quantization unit, 8... Inverse SVD orthogonal transform unit, 10.

Claims (3)

A prediction error between a block of the current image and a plurality of predetermined candidate prediction image blocks of the previous image is subjected to SVD conversion using SVD obtained from the candidate prediction image block, and the result of the SVD conversion is quantized / entropy coded An SVD motion detection unit for obtaining an SVD conversion base and a motion vector MV for the candidate prediction image block having the smallest code amount,
An SVD orthogonal transform unit that orthogonally transforms a prediction error, which is a difference between a current image and a predicted image motion-compensated with the motion vector MV, using the SVD transform base obtained by the SVD motion detection unit;
A motion compensated prediction singular value expansion encoding apparatus comprising: means for quantizing and entropy encoding a conversion result by the SVD orthogonal transform unit. The SVD conversion for a prediction error between a block of the current image and a plurality of predetermined candidate prediction image blocks of the previous image is performed using an SVD conversion base of the candidate prediction image block. Item 2. The motion compensated prediction singular value expansion coding apparatus according to Item 1. Means for dequantizing the data quantized by the quantization means;
Inverse SVD orthogonal transform means for performing inverse SVD orthogonal transform using the SVD transform base used in the SVD orthogonal transform unit;
Adding means for adding the motion compensated predicted image and the inverse SVD orthogonal transformed data;
The motion compensated prediction singular value expansion coding apparatus according to claim 1, wherein the previous image is generated.

Images