JP3907505B2 - Block truncation coding apparatus and method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a block truncation encoding apparatus and method , and more particularly to an apparatus that improves image quality by setting appropriate sections of resolution components.
[0002]
[Prior art]
Conventionally, a method called block truncation coding (BTC Block Truncation Coding) is known as a still image coding method. This means that the original image is decomposed into blocks of m × n pixels (division is not essential), a predetermined number of gradations is assigned to the block, and the pixel values in the block are expressed by the limited number of gradations. This is a method of (quantizing the number of gradations).
[0003]
FIG. 20 shows an outline of setting a quantization interval and a decoded value when a monochrome image is encoded by assigning four gradations. A thick line in FIG. 20 shows a histogram of pixel values of a block to be encoded, with the horizontal axis representing pixel values and the vertical axis representing frequency.
Based on the histogram, an average value LA of the pixel values in the block and a value LD reflecting the dynamic range of the pixel values in the block are obtained, and the pixel values are quantized based on these two quantities. Four sections are set. A flag indicating which section each pixel of the original image is included is called φ, and at the time of encoding, three quantities of LA, LD, and φ of each pixel are encoded. Note that LD is called a gradation width index, and φ is called a resolution component.
[0004]
Now, it is considered that the monochrome image (with a gradation of 8 bits per pixel) is encoded by being compressed to 1/2 by 2 × 4 pixels (64 bits in total) as one block. In order to assign 4 gradations to each block, 2 bits are required as φ per pixel, so a total of 16 bits are required with φ alone, and in order to compress to 1/2 (= 32 bits per block) The remaining 16 bits are assigned to LA and LD.
Here, the values of LA, LD, and φ are obtained as follows in general block truncation coding.
As shown in FIG. 20, the maximum value of the pixel values in the block is max, the minimum value is min, and the section (max-min) is equally divided into four. Then, average values Q1 and Q4 of the pixel values in each section are obtained for both ends of the four sections. And
LA = (Q1 + Q4) / 2
LD = Q4−Q1
, LA and LD are obtained, and based on this, the boundaries L1 and L2 of the quantization interval are
L1 = LA −LD / 4
L2 = LA + LD / 4
Is required. The important point here is that the quantization interval of the pixel value is set by dividing the LD into four equal parts. This is nothing but to quantize the vicinity of the median of the histogram finely and coarsely quantize the edges, and set the quantization interval assuming a histogram with a relatively high frequency of appearance near the median. It is thought that there is. This is because by densely quantizing a section having a high appearance frequency, the quantization error of the entire block can be suppressed to be small, and the image quality after decoding can be improved. In this way, assuming a general histogram, the BTC for setting the LD section is called a GBTC (Generalized Block Truncation Coding) type coding system, as described in JP-A-6-164950. .
Here, when x is a pixel value, φ is
When min ≤ x ≤ L1, φ = 0
If L1 <x ≦ LA φ = 1
When LA <x ≤ L2, φ = 2
When L2 <x ≤ max φ = 3
It is prescribed.
On the other hand, the decoded value d is
When φ = 0 d = d1 = Q1 (= LA-LD / 2)
When φ = 1 d = d2 = LA − LD / 6
When φ = 2 d = d3 = LA + LD / 6
When φ = 3 d = d4 = Q4 (= LA + LD / 2)
It is prescribed. What is important here is that the LD is divided into six equal parts to define the decoded values d2 and d3. This is nothing but decoding at a position farther from the center of each quantization interval as seen from LA, but this seems to intend to increase the dynamic range of the decoded value.
In the above, in order to quantize to 4 gradations, the section (max−min) is divided into 4 equal parts, and then LD is obtained. However, when quantizing to other gradations, K. Ogura; see "Generalized Block Truncation Coding (GBTC)", ISO / TC97 / SC2 / WG8 N510 (June 1987) (GBTC-type coding is a subset of GBTC).
[0005]
As described above, in the GBTC coding method, it is considered that the quantization interval is set assuming a histogram with a relatively high appearance frequency near the median, but the actual image may not be so. . For example, in the case of a monochrome character image as shown in FIG. 21, the histogram of the pixel values is as shown in FIG. 22, and the frequency at the center is low as opposed to FIG.
In the GBTC encoding method, 2 bits are uniformly assigned to all pixels in the block. However, when the original image is divided into small areas and encoded, the gradation of the target image itself is used. If is originally limited, uniform allocation is not efficient. That is, it is necessary to increase the allocation to φ when the gradation of the original image is wide, and to decrease the allocation to φ when it is narrow. (In Japanese Patent Application No. 2000-092393, “divides an image into rectangular blocks of a predetermined size, and further divides the block to create an image block as a unit of frequency conversion. (A fixed-length encoding device that originally encodes a rectangular block to a predetermined code length. ”Is intended for such a method.)
[0006]
Therefore, for example, the above 2 × 4 pixel block is divided into two 2 × 2 pixel sub-blocks, and the two sub-blocks are classified into “sub-blocks with narrow gradation” and “sub-blocks with wide gradation”. Consider assigning 1 gradation to the former and assigning 4 gradations to the latter. Thus, when viewed in blocks of 2 × 4 pixels, a total of 5 gradations (or more) can be obtained, and the image quality of the decoded image can be improved. In this case, a configuration in which only a representative value of a pixel is encoded for a “subblock with a narrow gradation” and BTC is applied to a “subblock with a wide gradation” can be considered.
Also in this case, the histogram of the pixel value of only “sub-block with wide gradation” is no longer the histogram of the whole block, and even if it is not a character image, it is different from the histogram assumed by the GBTC coding method. It will be a thing. In the first place, a “sub-block with a narrow gradation” is a sub-block with a small difference in pixel values within a sub-block, and a “sub-block with a wide gradation” has a large or small pixel value within a sub-block. It is a sub-block containing both things. Therefore, the pixel value histogram of only the “sub-block with wide gradation” is flat as shown in FIG. 23, or as shown in FIG. The shape may be changed.
[0007]
In such a case, since the assumption for the histogram does not hold, the GBTC coding method cannot be applied as it is. In order to obtain the optimum image quality, a new LD division method is essential. For example, as shown in FIG. 1, it is necessary to roughly divide the center of the LD and to densely divide both ends. Further, it is desirable that the decoded value is set from both ends of the gradation width index rather than the center of each section set as the resolution component.
[0008]
[Problems to be solved by the invention]
Therefore, an object of the present invention is to improve image quality by setting appropriate sections when applying block truncation coding to an image that may have a histogram that is not of a normal shape.
[0026]
Assuming a flat distribution as shown in FIG. 23 as a “non-regular histogram”, the section (max−min) divided into four equal parts is used as it is as a section of φ, and at the center of each section. By decoding, the quantization error can be minimized. This corresponds to dividing the LD into 1: 2: 2: 1 .
[0027]
Therefore, in addition to the above object , the present invention aims to minimize the quantization error in the case of a flat distribution.
[0035]
[Means for Solving the Problems]
In view of the above, the invention according to claim 1 is characterized in that the gradation width index is divided into three or more, and both end sides are divided more densely than the center of the gradation width index to set the resolution component section. A block truncation coding apparatus is proposed.
[0048]
The invention according to claim 2, a block truncation coding apparatus according to claim 1, the tone width index 1: 2: 2: Proposal block truncation coding apparatus characterized by divided into 1 To do.
[0050]
The invention according to claim 3 includes the step of dividing the gradation width index into three or more and dividing the both ends from the center of the gradation width index more densely to set the resolution component section. A block truncation coding method is proposed.
[0056]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, a preferred example of an embodiment of the present invention will be described with reference to the drawings. First, an apparatus configuration of the present embodiment will be described.
[0057]
[Device Configuration 1]
FIG. 2 shows a first example of a device configuration according to the present application. An HDD, RAM, and CPU are connected via a data bus, and an original image is encoded in the following flow.
{Circle around (1)} The original image recorded on the HDD is read onto the RAM by a command from the CPU.
(2) [Encoding step] The CPU partially reads an image on the RAM and applies the encoding method of the present application to perform encoding.
(3) The CPU writes the encoded data in another area on the RAM.
(4) When all the original images are encoded, the encoded data is recorded on the HDD in accordance with a command from the CPU.
In the same apparatus configuration, the encoded image is decoded according to the following flow.
(1) The encoded image recorded on the HDD is read onto the RAM in accordance with a command from the CPU.
(2) [Decoding step] The CPU partially reads the encoded image on the RAM and performs decoding by applying the present application.
(3) The CPU writes the decrypted data in another area on the RAM.
(4) When all the images are decoded, the decoded data is recorded on the HDD in accordance with a command from the CPU.
[0058]
[Device configuration 2]
FIG. 3 shows a second example of the device configuration according to the present application. The HDD, RAM 1 (inside PC), CPU 1 (inside PC), and printer are connected via a data bus. When the original image is printed out, the image is encoded, and the encoded data is transmitted to the printer. Since the amount of data transmitted to the printer is reduced, the transmission time is shortened, and high-speed printing is possible even when the time required for encoding / decoding is taken into account.
{Circle around (1)} The original image recorded on the HDD is read onto the RAM by a command from the CPU.
(2) [Encoding step] The CPU 1 partially reads an image on the RAM 1 and performs encoding by applying the encoding method of the present application.
(3) The CPU 1 writes the encoded data in another area on the RAM 1.
(4) The encoded data is recorded on the RAM 2 in the printer in accordance with a command from the CPU 1.
(5) [Decoding step] The CPU 2 in the printer reads the encoded data and applies the decoding process of the present application to decode the image.
(6) The CPU 2 writes the decrypted data on the RAM 2.
After all data is decrypted, the printer prints out the decrypted data according to a predetermined procedure.
With the above apparatus configuration, in the first embodiment of the present invention, a color original image (1 pixel 32 bits) composed of 4 components of CMYK is divided into blocks of 2 × 2 pixels and quantized into 4 gradations. Block truncation encoding is performed (FIG. 4). In this embodiment , the code configuration is shown in FIG. 5 (φ is 8 bits because bits of 4 pixels are connected).
Here, FIG. 7 shows processing for searching for the maximum value / minimum value within the block, FIG. 8 shows LA and LD calculation processing, and FIG. 9 shows φ calculation processing, which is common to the present embodiment. In the present application, as shown in FIG. 9, the LD is divided by 6 to densely quantize the both ends.
FIG. 10 shows the flow of the first embodiment of the present invention. First, after the pixel value P (i) of the C component (C plane) is input, the maximum value / minimum value in the block is obtained. When the maximum value-minimum value (which is an amount that directly reflects the distribution width of the pixel value) is equal to or greater than the predetermined value Th, BTC is applied as flag = 1, otherwise, flag = 0 is set. Encode the average of the pixel values in the block and repeat this for all components and blocks.
In this example, the pixel value of each component (plane) is 8 bits, and since LA, LD, and average value are all assigned 8 bits as shown in FIG. 5, LA, LD, and average value are not quantized. Can be used as they are. In this example, for each block, it is determined whether BTC is applied or the average value is encoded, and this is performed for all the planes. Therefore, the arrangement of codes is as shown in FIG. 6, for example.
When decoding these codes, first, the flag is read, and when encoded with BTC, LA, LD, and φ are read, and the decoded value d is
When φ = 0 d = d1 = Q1 (= LA-LD / 2)
When φ = 1 d = d2 = LA − LD / 6
When φ = 2 d = d3 = LA + LD / 6
When φ = 3 d = d4 = Q4 (= LA + LD / 2)
Get by. This is because, as described previously, but normally decoded value means to select the from both ends of the quantization intervals, particularly when the histogram is flat, means to decode the central segments. In addition, when it is not encoded by BTC, the code becomes a decoded value for four pixels as it is (note that the flowchart is omitted in this specification ).
[0060]
Further, FIG. 14 shows the flow of the present embodiment 2. In this example, a color original image (24 bits per pixel) composed of 3 components of CMY is divided into blocks of 2 × 2 pixel units, quantized into 4 gradations, and block truncation coding is performed. First, after the pixel value P (i) of the M component is input, the maximum value / minimum value in the block is obtained, and the maximum value-minimum value (this is a quantity that directly reflects the distribution width of the pixel value). When the value is equal to or greater than the predetermined value Th, the BTC is applied with the flag = 1, and when not, the average value of the pixel values in the block is encoded with the flag = 0. Next, regarding other components, the result of the M component is reflected as it is, and it is determined whether or not to use BTC. Therefore, the code is as shown in FIG.
[0061]
Further, FIG. 15 shows the flow of the present embodiment 3. In this example, a color original image (24 bits per pixel) composed of three RGB components is divided into blocks of 2 × 2 pixel units, quantized into four gradations, and block truncation coding is performed. First, all pixel values in the block are converted into luminance Y and color differences U and V by the following formula. This is a transformation called RCT (Reversible Component Transform).
Luminance Y = | _ (R + 2G + B) / 4_ | (symbol | __ | represents a floor function)
Color difference U = RG-Formula 1
Color difference V = BG
Inverse transformation is
R = G + U
G = Y− | _ (U + V) / 4_ |
B = V + G
Then, the S conversion shown in FIG. 13 is performed on the luminance (a, b, c, d) for four pixels in the block, and four coefficients LL, HL, LH, and HH are generated. The S conversion is a simple wavelet filter, and the HL, LH, and HH coefficients correspond to the high-pass filter output. If any of these values is equal to or greater than a predetermined value Th (HH is 2T), BTC is applied to all of YUV, and if not, an average value is encoded. Therefore, the code is as shown in FIG.
When decoding these codes, first, the flag is read, and when encoded with BTC, LA, LD, φ are read, and the decoded value d is
When φ = 0 d = d1 = Q1 (= LA-LD / 2)
When φ = 1 d = d2 = LA − LD / 6
When φ = 2 d = d3 = LA + LD / 6
When φ = 3 d = d4 = Q4 (= LA + LD / 2)
Get by. In addition, when the encoding is not performed by BTC, the code is a decoded value for four pixels as it is. Thereafter, the pixel value can be obtained by returning YUV to RGB by the above equation 2 (the flowchart is omitted in this specification ).
[0062]
In Embodiment 4 of the present application, FIG. 16 divides a color original image (1 pixel 24 bits) composed of 3 components of RGB into blocks of 2 × 4 pixels and further sub-blocks A and B of 2 × 2 pixels. Fig. 2 shows a state in which re-division and encoding are performed. This example illustrates the code structure in Figure 17.
FIG. 19 shows the flow. The RGB value of each sub-block is YUV converted, and thereafter the luminance or color difference is S-converted. The suitability of the BTC is determined by the relationship between the S conversion coefficient and the predetermined value.
Both sub-blocks A and B are encoded with an average value. Flag = 0
Sub-block A is the average value, B is BTC. Flag = 1 and φ is 8bit
Sub-block A is BTC, B is average value. Flag = 2 and φ is 8bit
Sub-blocks A and B are combined into 8 pixels for BTC. Flag = 3 and φ is 16bit
(LA and LD are divided by 2 and quantized to 7 bits)
This is repeated for all luminance and color differences and all blocks. Therefore, an example of the code is as shown in FIG. Rather than applying BTC individually to each sub-block as in the first embodiment, applying BTC together in two sub-blocks as in this example reduces the amount of LA and LD required by one set. You can raise the rate.
Similarly, in the case of decoding, first, the flag is read, and in the case of encoding with BTC, LA, LD, and φ are read, and the decoded value d is set as
When φ = 0 d = d1 = Q1 (= LA-LD / 2)
When φ = 1 d = d2 = LA − LD / 6
When φ = 2 d = d3 = LA + LD / 6
When φ = 3 d = d4 = Q4 (= LA + LD / 2)
Get by. If the average value is also encoded, the code becomes the decoded value for four pixels as it is. Thereafter, the pixel value can be obtained by returning YUV to RGB by the above equation 2 (the flowchart is omitted in this specification ).
[0064]
【The invention's effect】
Since the block truncation encoding apparatus according to claim 1 divides the gradation width index into three or more, and finely divides both ends of the gradation width index from the center to set the resolution component section, When block truncation coding is applied to an image that may have a histogram that is not the shape of the image, it is possible to improve image quality by setting an appropriate section.
[0077]
Block truncation coding apparatus according to claim 2 is a block truncation coding apparatus according to claim 1, the tone width index 1: 2: 2: Since is divided into 1, when the flat distribution In addition, the quantization error can be minimized.
[0079]
The block truncation encoding method according to claim 3 includes a step of dividing the gradation width index into three or more and dividing the both ends from the center of the gradation width index more densely to set a resolution component section. Therefore, in the case where block truncation coding is applied to an image that may have a histogram that does not have a normal shape, image quality can be improved by setting an appropriate section.
[Brief description of the drawings]
FIG. 1 is a section setting method of the present application.
FIG. 2 is a diagram of an apparatus configuration example 1 of the present application.
FIG. 3 is a diagram of an apparatus configuration example 2 of the present application.
FIG. 4 is a diagram in which a color original image (1 pixel 32 bits) composed of four CMYK components according to Embodiment 1 of the present application is divided into blocks of 2 × 2 pixel units.
FIG. 5 is a code configuration diagram of Embodiment 1 of the present application.
FIG. 6 is an example of an arrangement of codes according to the first embodiment of the present application.
FIG. 7 is a flowchart of maximum / minimum value search processing according to the embodiment of the present invention.
FIG. 8 is a flowchart of LA and LD calculation processing according to the embodiment of the present invention.
FIG. 9 is a flowchart of φ calculation processing according to the embodiment of the present application.
FIG. 10 is a flowchart of Embodiment 1 of the present application.
FIG. 11 is a code configuration diagram of Embodiment 2 of the present application.
FIG. 12 is a code configuration diagram of Embodiment 3 of the present application.
FIG. 13 is a conversion formula of S conversion and inverse S conversion.
FIG. 14 is a flowchart of Embodiment 2 of the present application.
FIG. 15 is a flowchart of Embodiment 3 of the present application.
FIG. 16 is a diagram in which a color original image (one pixel 24 bits) composed of three RGB components according to the fourth embodiment of the present application is divided into blocks of 2 × 4 pixel units.
FIG. 17 is a code configuration diagram of the fourth embodiment of the present application.
FIG. 18 is an example of an arrangement of codes according to the fourth embodiment of the present invention.
FIG. 19 is a flowchart of Embodiment 4 of the present application.
FIG. 20 is a schematic diagram of a method for setting each quantum interval and a decoded value in GBTC of the prior art.
FIG. 21 is an example of a monochrome character image according to a GBTC type encoding method of the prior art.
22 is a histogram of pixel values for the image of FIG.
FIG. 23 is an example of a histogram of pixel values of only “sub-block with wide gradation” in the GBTC type encoding method;
FIG. 24 is an example of a histogram of pixel values of only “sub-block with wide gradation” in the GBTC type encoding method;

Claims (3)

A block truncation coding apparatus characterized in that a gradation width index is divided into three or more, and both end sides are densely divided from the center of the gradation width index to set a resolution component section. A block truncation coding apparatus according to claim 1, the tone width index 1: 2: 2: block truncation coding apparatus characterized by divided into 1. A block truncation encoding method comprising the steps of: dividing a gradation width index into three or more, and dividing both ends of the gradation width index densely from the center to set a resolution component section.

Images