JPH042275A - Video signal decoding and reproducing device - Google Patents

Abstract

PURPOSE:To surely eliminate distortion between blocks by adding a random noise to a picture element data based on the activity calculated for each block. CONSTITUTION:When a compressed picture data is subject to Huffman, inverse normalizing and 2-dimension orthogonal conversion at a Huffman decoding section 12, an inverse normalizing section 14 and a 2dimension orthogonal conversion section 16 respectively, a block activity calculation section 36 takes notice of a specific block to calculate the activity. An activity discrimination section 38 compares the calculated activity with a threshold level and an output from the activity discrimination section 38 is sent to a low pass filter section 28 and a coefficient setting circuit 32. when the activity is low, a picture element at a block border by the low pass filter is corrected and random noise is added. When the activity of the picture is high, the correction of picture element and the addition of random noise are not implemented. Thus, a reproduced picture data whose inter-block distortion is reduced is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an image signal decoding and reproducing apparatus, and more particularly to an image signal decoding and reproducing apparatus that decodes and reproduces compression-encoded image data while reducing block distortion.

BACKGROUND ART When storing digital image data such as image data taken by an electronic still camera in a memory or transmitting it, various types of compression encoding are performed to reduce the amount of data. In particular, two-dimensional orthogonal transform encoding
It is widely used because it can perform encoding at a high compression rate and also suppress image distortion caused by encoding.

In such two-dimensional orthogonal transform encoding, image data is divided into a predetermined number of blocks, and the image data within each block is subjected to two-dimensional orthogonal transform. The orthogonally transformed image data, that is, the transformation coefficients, are compared with a predetermined threshold, and portions below the threshold are truncated (coefficient truncation). The transform coefficients whose coefficients have been truncated are divided by a predetermined quantization step value, that is, a normalization coefficient, and quantization, that is, normalization, is performed according to the step width. The normalized transform coefficients are then Huffman encoded to obtain compressed image data.

When decoding such compressed image data, after Huffman decoding, inverse normalization and orthogonal inverse transformation are performed to obtain the original image data. However, since the image data is compressed and encoded into blocks, orthogonally transformed and encoded, distortion occurs between blocks in the decoded image data.

Particularly in areas of a gentle image with little change, there is a problem in that the boundaries (edges) between blocks are noticeable and the image deteriorates.

Therefore, in order to reduce such distortion, it is conceivable to use a low-pass filter to soften the edges that appear at the boundaries of blocks, that is, the contours of blocks.

However, there is a problem in that the sharpness of the reproduced image deteriorates, and when a low-pass filter is used as described above, the high-frequency components are cut by the low-pass filter, which further deteriorates the sharpness of the reproduced image. Ta.

OBJECT OF THE INVENTION It is an object of the present invention to provide an image signal decoding and reproducing apparatus that can solve the problems of the prior art and obtain reproduced image data with a reduced number of block listeners of compressed and encoded image data. purpose.

According to the present invention, an image signal decoding and reproducing apparatus that decodes and reproduces digital image data that has been divided into a plurality of blocks and subjected to two-dimensional orthogonal transform encoding for the image data of each block is a digital A denormalization means for denormalizing the image data, and a denormalization means for denormalizing the data by the denormalization means.
orthogonal inverse transform means for performing dimensional orthogonal inverse transform; activity calculation means for calculating activity for each divided block of digital image data encoded by two-dimensional orthogonal transform; random noise generation means for generating random noise; Random noise addition means adds random noise generated from the random noise generation means to pixel data subjected to two-dimensional orthogonal inverse transformation by the inverse transformation means, and the random noise addition means adds random noise generated from the random noise generation means to pixel data subjected to two-dimensional orthogonal inverse transformation by the inverse transformation means, and the random noise addition means This adds random noise.

DESCRIPTION OF EMBODIMENTS Next, embodiments of the image signal decoding and reproducing apparatus according to the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 shows an embodiment of an image signal decoding and reproducing apparatus according to the present invention.

Compression-encoded image data is input to this device and decoded and reproduced. The image data decoded and reproduced by this apparatus is image data compressed and encoded by a compression encoding apparatus as shown in FIG. In the apparatus shown in FIG. 2, image data input through an input terminal 40 is divided into a plurality of blocks each consisting of 8x8 pixels in a block division section 42, and subjected to two-dimensional orthogonal transformation in a two-dimensional orthogonal transformation section 44. Further, the normalization unit 46 performs normalization by dividing by a normalization coefficient. The normalized data is Huffman encoded in a Huffman encoder 48.

The Huffman encoding unit 48 stores the normalized data in the fourth
As shown in the figure, the data is scanned in a zigzag pattern and input. Since zeros are often continuous in normalized conversion coefficients, we detect the continuous amount of zero value data, that is, the zero run length, find the zero run length and non-zero amplitude, and divide this into two. Dimensional Huffman encoding. Two-dimensional Huffman encoded image data is output from the output terminal 50.

The image data compressed and encoded in this manner is input to the apparatus shown in FIG. 1 and decoded. The decoding and reproducing apparatus shown in FIG. 1 has a Huffman decoding section I2. Compression-encoded image data is input to the Huffman decoding unit 12 from an input terminal IO, and is Huffman-decoded. The Huffman-decoded image data is sent to the denormalization unit 14 and denormalized. That is, it is multiplied by the normalization coefficient used for normalization in the apparatus of FIG. The image data denormalized by the denormalization unit 14 is sent to a two-dimensional orthogonal inverse transformation unit 16, where it is subjected to two-dimensional orthogonal inverse transformation. The image data subjected to orthogonal inverse transformation in the two-dimensional orthogonal inverse transformation unit 16 is sent to the pro-cn combination unit 18, where a plurality of blocks are combined and original image data is obtained.

The image data whose blocks have been synthesized by the block synthesis section 18 is sent to the low-pass filter correction section 20.

As will be described later, when the activity of a block is below a certain value, the low-pass filter correction unit 20 corrects pixel data near the block boundary using a low-pass filter, and corrects inter-block distortion, that is, block Performs processing to remove contours that appear on the boundaries of .

The output from the low-pass filter correction section 20 is sent to the random noise addition section 22. Random noise adding section 22
is an adder, which adds random noise sent from the multiplier 30 to the pixel data according to the activity, as will be described later. The output from the random noise adding section 22 is sent to the limiter section 26. The limiter unit 26 clips the pixel data to which random noise has been added in the random noise adding unit 22 using a predetermined threshold value. That is, when the level of pixel data deviates from a predetermined range, for example 0 to 255, due to the addition of random noise, the deviated portion is cut so that the pixel data falls within the range.

The output from the limiter section 26 is output to the output terminal 24.6 The image data subjected to the two-dimensional orthogonal inverse transformation in the two-dimensional orthogonal inverse transformation section 16 is sent to the block activity calculation section 36. The block activity calculation unit 36 calculates the activity for each block, that is, the degree to which the block contains image data of high frequency components, from the image data for each block that has been subjected to two-dimensional orthogonal inverse transformation.

The output from the block activity calculation unit 36 is sent to the activity determination unit 38.

Note that instead of calculating the activity of the block in the block activity calculation unit 36, the level of the activity of the block is determined by counting the amount of encoded data of compressed and encoded image data input from the input terminal IO. You may. The amount of encoded data increases in blocks with high activity, ie, blocks with many high frequency components, and decreases in blocks with low activity, ie, blocks with many low frequency components. In order to count the encoded data amount in this way, an encoded data amount counter connected to the input terminal 10 may be provided to count the encoded data amount and send its output to the activity determination section 38. good. The activity determination unit 38 compares the activity of the block calculated by the block activity calculation unit 36 with a predetermined threshold Th, and determines the level of the activity. The output from the activity determination section 38 is sent to the low-pass filter selection section 28 and the coefficient setting section 34.

The low-pass filter selection section 28 selects a low-pass filter to be used in the low-pass filter correction section 20, depending on whether the activity of the block is lower than a predetermined value based on the input from the block activity determination section 38. As shown in FIG. 3, the low-pass filter used in the low-pass filter correction unit 20 is a filter that processes a total of four pixels, two pixels at the boundaries of each block, along the vertical boundaries of the blocks, for example. These four pixels X1. X2. X3. By passing X4 through the O-pass filter, pixels X2 and X3 are replaced with values expressed by the following equation.

X2 2K l-X 1 + K 2・X2+ to 3・X3
+41X4-(11X3=Kl-X4+K2-X3+K3-
X 2 +K 4 ·
If it is lower than , add 1 to the coefficient used in the above formula.
Set 4 to ~ as follows.

K I =2/8. K 2 =3/8. K:
#=2/8. K 4 = 1/8 By setting such a coefficient, pixels X2 . X
3 is corrected by the values of adjacent pixels, reducing distortion appearing in the outline of the block.

On the other hand, if the activity of the block is greater than or equal to the predetermined value Th, the coefficients 1 to 4 are set as follows.

Kl=0, K2=1, K3=0, K4=0 When such coefficients are set, pixel X2 .
X3 is left at its original value and no correction is made.

The multiplier 30 includes a random number generation circuit 32 and a coefficient setting circuit 3.
4 is connected. In this device, the random number generation circuit 32 generates a random number from among ±2 and ±1.0, and this random number is used as random noise. Coefficient setting circuit 34
sets the coefficient β=2 or β:0 depending on whether the activity of the block is lower than a predetermined value based on the input from the block activity determination unit 38, and outputs it to the multiplier 30. When β:2 is set, the random number sent from the random number generation circuit 32 is multiplied by 2 and sent to the random noise adding section 22. Random noise addition section 2
2 adds this random number sent from the multiplier 30 to the pixel data output from the low-pass filter correction section 20 as random noise. By adding such random noise, distortion caused by contours appearing at the boundaries of blocks can be alleviated.

The block distortion correction operation performed by one apparatus will be described with reference to the flowchart in FIG.

When the compressed image data is subjected to Huffman decoding, inverse normalization, and two-dimensional orthogonal inverse transformation in the Huffman decoding unit 12, inverse normalization unit 14, and two-dimensional orthogonal inverse transformation unit 16, respectively, the block activity calculation unit 36 specifies Pay attention to the block and calculate the activity +102+. The activity determination unit 38 sets the calculated activity to a threshold T.
h and determine whether it is smaller than the threshold Th (1
04) The output from the activity determination section 38 is sent to the low-pass filter selection section 28 and the coefficient setting circuit 32.

If the activity is smaller than the threshold Th, the low-pass filter selection unit 28 sets the coefficients from 1 to 4 to 1; 2, K.
2=3, K3=2, K4; set to 1 (106)
By using the low-pass filter with the coefficient K selected in this way, the low-pass filter correction unit 20 calculates the equation (11f21
The block boundary pixel X2. Replaces X3. Pixel X2. The pixel data with X3 replaced is sent to the random noise adding section 22.

When the activity is smaller than the threshold Th, the coefficient setting circuit 34 sets the coefficient β=2.108) Multiplier 30
Output to. The multiplier 30 multiplies the random number sent from the random number generation circuit 32 by 2 and outputs it to the random noise adding section 22. The random noise addition section 22 adds data obtained by multiplying a random number by 2, sent from the 1 multiplier 30, to the pixel data sent from the low-pass filter correction section 20. The pixel data to which random noise has been added is sent to the limiter section 26. In this embodiment, the limiter section 26 clips the portion of data exceeding this range so that the level of the pixel data falls within the range of 0 to 255. takes a value in the range of 0 to 255. Therefore, the pixel data output from the low-pass filter correction section 20 is also 0.
It takes values in the range of ~255. Random noise adding section 22
When random noise is added to this pixel data, the pixel data may exceed the range of 0 to 255, so that portion is cut by the limiter section 26.

If the activity is not smaller than the threshold Th in step 104, the low-pass filter selection unit 28 sets the coefficient Kl- to 4 to Kl=0, K2=1, K3=0, K4=0 f
llO) By using the low-pass filter with the coefficient K selected in this way, the low-pass filter correction unit 20 can satisfy the above equation (
11+21, pixel X2. Process X3. In this case, by selecting the coefficient K as described above, the pixel X2°x3 is output as is.

Pixel data from the low-pass filter correction section 20 is sent to a random noise addition section 22.

If the activity is not smaller than the threshold Th, the coefficient setting circuit 34 sets a coefficient β20 (fl12) and outputs it to the multiplier 30. The multiplier 30 multiplies the random number sent from the random number generation circuit 32 by 0, and the multiplier 30 multiplies the random number sent from the random number generation circuit 32 by 0.
Output to 2. Therefore, in this case, since no random number is output to the random noise adding section 22, the random noise adding section 22 does not add random noise to the pixel data sent from the low-pass filter correction section 20. Therefore, the pixel data sent to the limiter section 26 is not clipped.

According to this device, when the activity of a block is low as described above, pixels at the block boundary are corrected using a low-pass filter, and random noise is added. Therefore, in the case of an image with low activity and marked block contour distortion, the contour distortion can be removed by adding a low-pass filter and random noise.

On the other hand, in the case of an image with high activity, contour distortion of the block is not noticeable, so pixel correction using a low-pass filter and random noise addition are not performed. Therefore, sharpness does not deteriorate due to addition of a low-pass filter or random noise.

Note that in the above example, the block activity calculation unit 36
The activity of each block is calculated in , but the difference between the activities of adjacent blocks is calculated,
Based on this value, the block listener may be corrected by adding a low-pass filter or random noise.

When the difference in the degree of low-frequency components and high-frequency components contained between adjacent blocks is small, distortion at the block boundaries becomes noticeable, so block listener processing is performed by adding a low-pass filter or random noise. , when the difference is large, the distortion that occurs at the block boundaries is not noticeable, so block listener processing is not performed.

Note that the block listener correction by the low-pass filter correction section 20 may be omitted, and only the random noise addition by the random noise addition section 22 may be performed.

Effects According to the present invention, random noise is added to one pixel data based on the activity calculated for each block. Therefore, for images with low activity and noticeable distortion between blocks, block listeners must be removed without fail, while for images with high activity and distortion between blocks not noticeable, random noise should not be added. This can prevent deterioration of sharpness.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of an image signal decoding and reproducing device according to the present invention, and FIG. 2 is a block diagram showing an example of an image signal compression encoding device.
FIG. 3 is a diagram showing pixel data at block boundaries, FIG. 4 is a diagram showing the order of Huffman encoding, and FIG. 5 is a diagram showing inter-block distortion correction operation. Description of Main Parts No. 8 Huffman decoding unit Inverse normalization unit Two-dimensional orthogonal inverse transformation unit Block synthesis unit Low-pass filter correction unit Random noise addition unit Low-pass filter selection unit Multiplier Random number generation circuit Block activity Calatabe Activity determination unit

Claims (1)

[Scope of Claims] 1. An image signal decoding and reproducing device that decodes and reproduces digital image data that is divided into a plurality of blocks and subjected to two-dimensional orthogonal transform encoding on the image data of each block, the device comprising: denormalization means for denormalizing digital image data; orthogonal inverse transformation means for performing two-dimensional orthogonal inverse transformation on the data denormalized by the denormalization means; and the two-dimensional orthogonal transformation-encoded digital image. an activity calculation means for calculating an activity for each divided block of data; a random noise generation means for generating random noise; an image signal decoding and reproducing apparatus, further comprising a random noise adding means for adding random noise generated from the block, wherein the random noise adding means adds the random noise based on the activity of each block. . 2. The device according to claim 1, further comprising: low-pass filter correction means for processing pixel data at the boundary portion of the block with a low-pass filter based on the activity for each block, and the random noise. An image signal decoding and reproducing apparatus, wherein the addition means adds the random noise to the pixel data processed by the low-pass filter correction means. 3. The device according to claim 1 or 2, wherein the random noise adding means compares the activity of each block with a predetermined threshold, and adds the random noise only to blocks in which the activity is smaller than the threshold. An image signal decoding and reproducing device characterized by performing the following.