JP2614661B2 - High efficiency image coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention predicts the occurrence of mosquito noise, varies the quantization step size of a quantizer, and enables high-efficiency encoding with little mosquito noise. The present invention relates to an efficiency image coding device.

[Prior Art] Since image data has a large amount of information, high-efficiency encoding for reducing the average number of bits per pixel is effective for transmission or recording on a storage medium. High-efficiency coding includes predictive coding, transform coding, and other variable-length coding.
Various coding methods are used, and these various coding methods are often used in combination.

A quantizer used in a high-efficiency image coding apparatus usually performs nonlinear quantization. At this time, the quantization step size selects a quantization step size suitable for the number of remaining bits in the buffer memory or the prediction error amount. Is common. Also, in the case of a high-efficiency image coding apparatus using orthogonal transform, the screen is divided into blocks each having an appropriate number of pixels, and processing is performed for each block. Alternatively, one frame is divided into strip-shaped areas, and this area is used as a unit.

[Problems to be Solved by the Invention] In a conventional high-efficiency image encoding apparatus that uses a combination of predictive encoding and orthogonal transform, when coarse quantization is performed in a quantizer, a decoded image is degraded by quantization noise. In particular, in the case of a block that has both an edge portion and a flat portion of an image, the effect of blocking the high-frequency component of the edge portion spreads to the flat portion, and the degradation of the decoded image is very remarkable. Met. The edge referred to here means a solid edge having a finite width, a portion where a concentration gradient exists, and essentially indicates a region. Such degradation of the decoded image is called mosquito noise because when the original image is a human image, it appears concentrated at the boundary between the human image and the background, and mosquitoes appear to be swarming around the human image.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a high-efficiency image encoding device that controls the quantization characteristics of a quantizer for each block and prevents the occurrence of mosquito noise.

[Means for Solving the Problems] In order to achieve the above object, a high-efficiency image encoding device according to the present invention provides a prediction error obtained by predictively encoding image data for each block in units of several pixels. An orthogonal transformer for orthogonally transforming, a quantizer for quantizing the prediction error orthogonally transformed by the orthogonal transformer in accordance with a quantization step set externally, and the image data adjacent to each block. Pixels are subjected to difference processing, individual pixels are classified into flat, intermediate, and edge groups based on the obtained density gradient, and a block in which mosquito noise is predicted to be generated is determined based on the number of pixels belonging to each group. A mosquito noise predictor, and performs the quantization step on a block determined to be a block in which mosquito noise occurrence is predicted more finely than other blocks. It is characterized by doing.

[Operation] A block in which mosquito noise occurs is a block in which both a flat region and an edge region coexist. Therefore, each pixel in the block is divided into flat, intermediate, and edge groups. Next, the number of pixels belonging to each group is counted, and it is determined whether or not the block is a mosquito noise generation block from the count values included in the flat, middle, and edge. This judgment is determined empirically.

For the mosquito noise generating block, the quantization step size is made smaller than the other blocks, and the block is finely quantized. As a result, high-frequency components of edges remain, and no mosquito noise appears in the image.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a diagram for explaining a mosquito noise prediction procedure by the high-efficiency image encoding device of the present invention. FIG. 2 is a block diagram showing one embodiment of the high-efficiency image encoding device of the present invention. FIG. 3 is a schematic configuration diagram showing each embodiment of the quantizer shown in FIG.

The high-efficiency image coding apparatus shown in FIG. 2 shows an embodiment in which transform coding (orthogonal transform), predictive coding, and variable-length coding are combined. However, in the circuit block shown in the figure, a circuit portion for distinguishing between-frame and intra-frame processing, a motion compensation circuit portion, and the like are omitted. 1 is
This is a noise filter that performs pre-processing of image data. 2
Is a mosquito noise predictor that predicts the occurrence of mosquito noise for each block. Reference numeral 3 denotes a subtractor that generates a difference signal from the previous frame block. Reference numeral 4 denotes a prediction error calculator that calculates an inter-frame prediction error amount. Reference numeral 5 denotes an orthogonal transformer (DCT) that linearly transforms image data into transform axes independent of each other for each block. Reference numeral 6 denotes a quantizer for setting the quantization step size externally. The output of the quantizer 6 is stored in a buffer memory 12 via a variable length encoder (VLC) 11. The data in the buffer memory 12 is output at an appropriate rate. A prediction signal is created by the inverse quantizer (DCT ^-1 ) 7, the inverse orthogonal transformer 8, the adder 9, and the frame memory 10, and the predicted signal is guided to the subtractor 3. The information of the quantization step size adopted by the quantizer 6 is guided to the inverse quantizer 7 as a signal 6a, and the inverse quantization is performed on the same block based on the same quantization step size as at the time of quantization. You.

The quantizer 6 has a configuration shown in FIG. Incidentally, the conventional quantizer includes a quantization characteristic control table 61 and a quantization circuit 60. The quantization characteristic control table 61 includes an rms (root mean square error) signal output from the prediction error meter 4 and a buffer. A BM signal indicating the number of remaining memory bits output from the memory 12 is supplied, and outputs an optimal quantization step size. The quantization circuit 60 performs quantization with a size according to this output.

In general, the image quality is improved by reducing the quantization step size, but the generated code amount increases. Conversely, if the quantization step size is increased, the generated code amount decreases, but the image quality deteriorates. Even with the same quantization step size, the generated code amount increases as the inter-frame prediction error amount increases. Therefore, in the quantization characteristic control table 61, an optimal quantization step size is selected from the number of remaining bits in the buffer memory 12 and the inter-frame prediction error amount so as to keep the generated code amount as constant as possible. In addition,
Since the high-efficiency image coding apparatus shown in FIG. 2 operates on a block basis, the prediction error weigher 4 has a frame memory therein, and measures the prediction error amount between frames for each frame. I have.

Although the above is the description of the conventional method, in this embodiment of the present invention, the mosquito noise predictor 2 is inserted after the noise filter 1 to perform prediction for each block of the input video signal, and the mosquito noise is predicted to be generated. About MN
A signal is generated and sent to the quantizer 6.

Specifically, the quantizer 6, as shown in FIG. 3 (a), a quantization characteristic control table 61, and Q _M 62 to set a finer step size, the switch 63 is switched by the MN signal And selectively connected to the quantization circuit 60 via the. Switch 63, when MN signal is at a high level, for switching to Q _M 62 side, with respect to the region where the occurrence of mosquito noise is expected, fine quantization Q _M 62 is held to the quantization circuit 60 The step size is set.

Incidentally, in the case of the above embodiment shown in FIG.
Although the quantization step size can be determined completely independently of the s signal and the BM signal, the step size is related to the rms signal and the BM signal as in a quantizer 6 shown in FIG. 3 (b). It is also possible to adopt a configuration for determining. In the case of this embodiment, the quantization characteristic control table 61 is directly connected to one switching terminal of the switch 63, and the quantization characteristic control table 61 is connected to the other switching terminal via Q _M '62'. It is. Q _M '62' functions to further finely set the quantization step size specified by the quantization characteristic control table 61 according to the rms signal and the BM signal by several steps.

Next, a procedure in which the mosquito noise predictor 2 predicts the occurrence of mosquito noise for each block will be described with reference to FIG. The procedure a is a procedure for calculating a level difference (density gradient) between the target pixel and the adjacent pixel. The mask 1 determines a horizontal difference S _m1 , and the mask 2 determines a vertical difference S _m2 . The concentration gradient is the square root of the sum of squares of S _m1 and S _m2 , but here, it is approximately treated as the sum of the absolute values of the two. Next, in the procedure b, the density gradient obtained in the previous procedure a is determined as a threshold based on the thresholds th1 and th2, and each pixel is determined to be flat, intermediate, or
Classify into each group of edges. After all the pixels belonging to the block have been subjected to group classification, the process proceeds to step c. In step c, the number of pixels belonging to each group is determined as a threshold based on the thresholds Th1, Th2, Th3, and Th4, and a block in which mosquito noise is predicted to occur is determined. Here, when all the conditions shown in the procedure c are satisfied, the block is set as a mosquito noise generating block.

Hereinafter, various parameters for estimating the mosquito noise occurrence block will be described. For example, if the signal level of the original image is in the range of 60-255,
The thresholds for grouping each of the middle and the edge are set to th1 = 20, th2 = 100. As the block determination condition of the procedure c, when the block size is 64 pixels with an aspect ratio of 8 × 8, Th1 = 20, Th2 = 20, Th3 = 3, Th4 = 32. That is, a block in which the flat area is 30% or more of the entire area, the intermediate area is 30% or less, and the edge area is between 5% and 50% is determined to be a mosquito noise generating block.

As described above, the high-efficiency image encoding apparatus orthogonally transforms a prediction error obtained by predictively encoding image data for each block in units of several pixels, and externally sets the orthogonally transformed prediction error. The mosquito noise estimator 2 for estimating the amount of mosquito noise generated around the edge of the image is different from the quantizer 6 for quantizing in accordance with the quantization step according to the quantization step. Because the configuration is such that the quantization step size is set finely, instead of variably controlling the quantization step size of the quantizer for each frame, the quantization step size is variably controlled for each block that divides the frame into several. Can be. In particular, since mosquito noise is predicted to occur in a block in which both a flat region and an edge region coexist, high-frequency components contained in the edge portion of the image are retained by setting a quantization step size finer than in other blocks. As a result, the generation of mosquito noise can be suppressed and a visually favorable image can be reproduced.

Furthermore, the mosquito noise predictor 2 performs difference processing on image data between adjacent pixels on a block-by-block basis, discriminates the obtained density gradient as a threshold, and flattens / intermediates / intermediates the individual pixels.
Edges are classified into each group, and a block in which mosquito noise is predicted is determined based on the number of pixels belonging to each group. For example, a flat area is 30% or more of the whole area, an intermediate area is 30% or less, and an edge area is 5% or less. By specifying a block between% and 50% as a block in which the occurrence of mosquito noise is predicted, and by setting the quantization step size finely, the occurrence of mosquito noise can be suppressed satisfactorily. , And high-efficiency image coding with high accuracy.

[Effects of the Invention] As described above, according to the present invention, a prediction error obtained by predictive coding of image data is orthogonally transformed for each block in units of several pixels, and the orthogonally transformed prediction error For a quantizer that quantizes according to a quantization step set externally, the mosquito noise predictor makes the quantization step size finer for blocks where mosquito noise is predicted than for other blocks, Rather than variably controlling the quantization step size of the quantizer on a frame basis, the quantization step size can be variably controlled for each block that divides the frame into several parts. Particularly, both the flat region and the edge region coexist. Since the occurrence of mosquito noise is predicted for each block, by setting the quantization step size finer than for other blocks,
High-frequency components contained in the edge portion of the image are retained, and the generation of mosquito noise is suppressed, so that a visually good image can be reproduced.Furthermore, the mosquito noise predictor converts the image data in block units into adjacent pixels. Differential processing is performed, the obtained density gradient is discriminated as a threshold value, individual pixels are classified into flat, intermediate, and edge groups, and a block in which mosquito noise occurrence is predicted is determined from the number of pixels belonging to each group. So, for example, the flat area is 30
%, The middle area is 30% or less, and the edge area is 5% to 50%.
By specifying the block in between as the block in which the occurrence of mosquito noise is predicted and setting the quantization step size finely, the occurrence of mosquito noise can be suppressed well, and high-precision It provides excellent effects such as high efficiency image coding.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a mosquito noise prediction procedure by the high-efficiency image encoding device of the present invention. FIG. 2 is a block diagram showing one embodiment of the high-efficiency image encoding device of the present invention. FIG. 3 is a schematic configuration diagram showing each embodiment of the quantizer shown in FIG. 2 mosquito noise generation predictor 4… (in-frame) prediction error metric 5… orthogonal transform 6… quantizer 11… variable length coder 12… buffer memory 60… quantizer 61 …… Quantization characteristic control table 62 …… Q _M 62 ′ …… Q _M ′ 63 …… Switch

Claims (1)

(57) [Claims] An orthogonal transformer for orthogonally transforming a prediction error obtained by predictive encoding of image data for each block in units of several pixels, and a prediction error orthogonally transformed by the orthogonal transformer, A quantizer that quantizes according to the set quantization step, and performs a difference process between the adjacent pixels on the image data in units of the blocks, and flattens, middles, and reduces the individual pixels based on the obtained density gradient. A mosquito noise predictor that classifies edges into groups and determines a block in which mosquito noise occurrence is predicted from the number of pixels belonging to each group; On the other hand, a high-efficiency image encoding apparatus characterized in that the quantization step is made finer than other blocks.