JP5270592B2 - Image coding apparatus, image coding method, and image coding program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce a code amount by taking the local property of images into consideration in bit depth conversion processing when encoding high bit depth images. <P>SOLUTION: A bit depth conversion processing part 10 divides the image of input image signals into a plurality of areas by an area division part 100 and converts the input image signals to low bit depth signals by divided area. At the time, the area division part 100 divides the area so as to minimize the sum of bit depth conversion errors. As encoded streams, division information encoded data, conversion table encoded data by area, low bit depth signal encoded data, and residual signal encoded data with the input image signals by the bit depth conversion are outputted. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

The present invention relates to encoding technology of image signals having a high dynamic range, techniques for particular reduce the amount of codes in the high bit depth video coding.

  In recent years, with the improvement of image quality, expectations for high dynamic range images are increasing. Accordingly, devices capable of acquiring a high bit depth signal in which the bit depth of the acquired signal is expanded from the conventional 8 bits to 10 bits or more have appeared. However, since the code amount of an image increases as the bit depth of the image signal increases, an efficient encoding method is necessary. Therefore, research on coding and decoding for high bit depth signals is underway.

  As a coding technique for a high bit depth signal, a technique as shown in FIG.

  In the method illustrated in FIG. 14, an image signal of a high bit depth signal (assumed to be an N bit signal) is input, and the bit depth conversion processing unit 300 performs a bit depth conversion process, whereby an N bit signal is (N−Δ). Convert bit signal to low bit depth signal. It is encoded by the encoding processing unit 301 and further decoded by the decoding processing unit 302. The inverse bit depth conversion processing unit 303 performs an inverse bit depth conversion process on the decoded image of the (N−Δ) bit signal to generate a high bit depth image of the N bit signal.

  Subsequently, the difference between the high bit depth image and the input signal is calculated by the subtracter 304, and the resulting residual signal is encoded by the encoder 305. The output is an encoded stream of residual signals and an encoded stream of low bit depth signals. Thus, this method is a method corresponding to scalable coding of bit depth.

  Also, instead of outputting the difference between the high bit signal image and the input signal, the image signal is input and converted to a low bit depth signal by performing bit depth conversion processing, and then encoded and decoded. There is also a method of performing reverse bit depth conversion processing on a low bit depth image and outputting a high bit depth image.

  The encoding efficiency in the method shown in FIG. 14 greatly depends on the bit depth conversion process. In this method, the amount of code of the output signal can be suppressed by suppressing the square sum (hereinafter referred to as bit depth conversion error) of the difference value between the input signal and the signal after the inverse bit depth conversion processing.

  In bit depth conversion processing, some techniques using Tone Mapping are proposed by Non-Patent Document 2 and the like.

  However, this conventional method is intended to maintain the subjective image quality of the image after bit depth conversion, and there is no guarantee that the bit depth conversion error is minimized, and the decoded signal may be deteriorated. Therefore, in Non-Patent Document 3, in order to design a bit depth conversion processing unit that minimizes a bit depth conversion error, a method for designing focusing on a histogram of pixel values of an input image is proposed as follows. Yes.

The bit depth of the input image is N bits, and the bit depth after the bit depth conversion process is (N−Δ) bits (Δ> 0). Considering the case of converting an N-bit image signal into an (N−Δ) -bit image signal, 2 ^N classes of the histogram are divided into 2 (N−Δ) power classes in the bit depth conversion process. Because of the processing, the concept of quantization is applied to the bit depth conversion processing. In addition, the bit depth conversion error can be reduced by optimizing the width of the divided section according to the distribution of the histogram, such as finely dividing at high frequency parts and coarsely dividing at low frequency parts.

  For the above reasons, Non-Patent Document 3 considers a technique in which an optimum quantization technique (Lloyd-Max quantization) is applied to bit depth conversion processing in order to minimize bit depth conversion errors. When converting an N-bit image signal to an (N−Δ) -bit image signal, a conversion table describing the relationship between the (N−Δ) -bit image signal, the class number, and the representative pixel value is obtained. It is done. This conversion table is used when reverse bit depth conversion processing is performed from an (N−Δ) -bit image signal to an N-bit image signal in the reverse bit depth conversion processing.

“Bit-Depth Scalable Video Coding”, M. Winken, D. Marpe, H. Schwarz, and T. Wiegand, Proc. IEEE Intl. Conf. On Images Processing (ICIP2007), 2007. “Photographic Tone Reproduction for Digital Images”, Erik Reinhard, Michael Stark, Peter Shirley, and Jim Ferwerda, In SIGGRAPH 2002 Conference Proceeding, ACM SIGGRAPH, Addison Wesley, pp.267-277, August 2002. “Basic study on coding of high bit depth image using bit depth conversion processing”, Ken Ito, Yukihiro Bando, Noriyuki Takamura, Hitoshi Uekura, Yoshiyuki Yashima, IEICE General Conference, 2009.

  In the conventional method, the bit depth conversion error for each frame can be reduced, but the locality within the frame cannot be taken into consideration. Since bit depth conversion is performed using a single conversion table even in parts with different textures, such as the sky and land, a bit depth conversion error may increase. Therefore, it is necessary to design a bit depth conversion process considering locality in the frame.

  The present invention has been made in view of such circumstances, and an object thereof is to further reduce the bit depth conversion error in comparison with the prior art in consideration of locality in a frame.

  In order to solve the above-described problems, the present invention has the main feature that the frame is divided into a plurality of regions at optimum division positions so as to minimize the bit depth conversion error. In the present invention, an image signal with a high bit depth is encoded using a bit depth conversion processing method for performing region division as described below.

[Adaptation of bit depth conversion processing]
In the present invention, in order to reduce the bit depth conversion error, a bit depth process for performing region division in a frame is proposed. A method for determining a division point for dividing an area is, for example, dividing a frame into two areas at each division point while moving the division point on the edge of the frame, and performing bit depth conversion processing / inverse bit depth in each area. The bit depth conversion error in each area is calculated by performing the conversion process. By setting the position where the sum of the bit depth conversion errors is minimum as the division point, it leads to reduction of the bit depth conversion error in the entire frame. In the case of the division number M, the bit depth conversion error for each region is calculated by performing bit depth conversion processing / inverse bit depth conversion processing in each region while moving M−1 division points. , M−1 division points are determined such that the sum of M bit depth conversion errors is minimized.

3 in divided over the division point decision, among the areas obtained by split, repeatedly applying the task of further divided into two regions bit depth conversion error is large.

[Reduction of additional information accompanying adaptation]
In order to reduce the code amount of the increased conversion table, encoding is performed using the correlation information in the table and between the tables. In the intra-table correlation, the fact that there is little difference from the previous element is used, the difference from the previous element is calculated, and the amount of information is reduced by encoding (intra-table difference). In the correlation between tables, the difference between the previous frame and the conversion table is small, the difference from the previous frame is calculated, and the amount of information is reduced by performing encoding (between tables). Difference).

  For these encodings, Golomb encoding, which is an encoding method suitable for an information source according to the exponential distribution, is applied. Furthermore, fixed-length encoding is performed with N bits, which is the number of bits of the input signal. Of the above three methods of intra-table difference, inter-table difference, and fixed-length encoding, the method that minimizes the code amount of the conversion table is applied to each region to reduce the additional information.

[Determination of optimal number of divisions]
By increasing the number of divisions, the code amount of the residual signal is reduced by reducing the bit depth conversion error, and the code amount of the table is increased by increasing the number of conversion tables. Accordingly, it is expected that the sum of the code amount of the residual signal and the code amount of the conversion table is minimized in a certain number of divisions. The division number that minimizes the sum of the code amounts is set as the optimum division number, and the present invention proposes a method of dividing the area into the optimum division number.

  According to the present invention, in the bit depth conversion processing when encoding a high bit depth image, it is possible to reduce the code amount by performing region division in consideration of local characteristics of the image.

It is a figure which shows the structural example of the image coding apparatus which concerns on one Embodiment of this invention. It is a flowchart of the encoding process by an image coding apparatus. It is a figure which shows the structural example of a bit depth conversion process part. It is a figure which shows the example of the area | region division method. It is a flowchart of a bit depth conversion process. It is a flowchart of another bit depth conversion process. It is a figure which shows the structural example of a conversion table encoding process part. It is a flowchart of a conversion table encoding process. It is a figure which shows the structural example of the image decoding apparatus relevant to this invention. It is a flowchart of the decoding process by an image decoding apparatus. It is a figure which shows the structural example of a conversion table decoding process part. It is a flowchart of a conversion table decoding process. It is a figure which shows the experimental result for effect verification. It is a figure which shows the structural example of the image coding apparatus using the conventional bit depth conversion.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[Image coding device]
FIG. 1 is a diagram illustrating a configuration example of an image encoding device according to an embodiment of the present invention. The image encoding device 1 converts an N-bit high bit depth signal into a low bit depth signal of (N−Δ) bits (Δ> 0), encodes the low bit depth signal, Encoding of a residual signal with a bit depth signal and encoding of a conversion table used for bit depth conversion processing are performed, and an encoded stream is output. The major difference from the prior art described in FIG. 14 is that the screen (frame) of the input image signal is divided into a plurality of areas, and the input image signal is reduced by using a conversion table created for each of the divided areas. Conversion to a bit depth signal is also to determine an optimal division position (and the number of divisions) in the screen so that a bit depth conversion error is minimized.

  The image encoding device 1 includes a bit depth conversion processing unit 10 that converts an N-bit input image signal into an (N−Δ) -bit low bit depth signal for each area obtained by dividing the screen, and a conversion used for bit depth conversion. A conversion table storage unit 11 for storing a table, a low bit depth signal encoding processing unit 12 for encoding a low bit depth signal of (N−Δ) bits, and a decoding process for decoding the encoded low bit depth signal Unit 13, an inverse bit depth conversion processing unit 14 that converts a (N−Δ) -bit decoded signal into an N-bit decoded signal, and a subtraction that calculates a difference between the N-bit input image signal and the N-bit decoded signal 15, a conversion table encoding processing unit 16 that encodes the conversion table, a residual signal encoding processing unit 17 that encodes the residual signal that is the output of the subtractor 15, and encoding region division information Split to And a broadcast encoding unit 18.

  Division information encoded data output from the division information encoding processing unit 18, conversion table encoded data output from the conversion table encoding processing unit 16, residual signal encoded data output from the residual signal encoding processing unit 17, The (N−Δ) bit signal encoded data output from the low bit depth signal encoding processing unit 12 is combined by a stream combining unit (not shown) and output as an encoded stream.

  The encoded stream output from the image encoding apparatus 1 is different from the conventional example shown in FIG. 14 in that the encoded stream includes low-bit depth signal encoded data and residual signal encoded data. In addition, encoded data of division information indicating the division position of the screen area is included for each frame, and encoded data of the conversion table is added to each divided area. The division information may include information on the number of divisions according to the embodiment.

  The bit depth conversion processing unit 10 includes an area dividing unit 100 that divides a screen into M (M ≧ 2) areas for bit depth conversion of an N-bit input image signal. In this example, the region dividing unit 100 is shown to be provided in the bit depth conversion processing unit 10. However, even if the region dividing unit 100 is provided as a separate module before the bit depth conversion processing unit 10, There is no problem in implementing the present invention.

  FIG. 2 is a flowchart of the encoding process performed by the image encoding device 1.

  In step S10, the bit depth conversion processing unit 10 divides the input image into M regions by the region dividing unit 100, and converts the input signal into a low bit depth signal while creating a conversion table for each region. The created conversion table is stored in the conversion table storage unit 11 with region information added if necessary. The division position of the area is selected so that the sum of the bit depth conversion errors in each area is minimized.

  In step S11, the division information encoding processing unit 18 encodes division information indicating how the region is divided. The division information may be encoded in combination with header information or other encoding parameter information.

  In step S12, the low bit depth signal encoding processing unit 12 encodes the (N−Δ) bit signal after the bit depth conversion. The (N−Δ) bit signal encoding process in the low bit depth signal encoding processing unit 12 is, for example, the conventional H.264 standard. Since it may be exactly the same as the processing of the encoder in the H.264 encoding method, a detailed description of the internal processing configuration will be omitted.

  In step S13, the decoding processing unit 13 decodes the encoded stream of the encoded (N−Δ) bit signal. Here, the decryption processing by the decryption processing unit 13 is, for example, conventional H.264. Since it may be exactly the same as the processing of the decoder in the H.264 encoding method, a detailed description of the internal processing configuration is omitted.

  In step S <b> 14, the inverse bit depth conversion processing unit 14 converts the (N−Δ) -bit decoded signal into an N-bit high bit depth signal using the conversion table stored in the conversion table storage unit 11.

  In step S15, the subtracter 15 calculates a difference between the high bit depth signal output from the inverse bit depth conversion processing unit 14 and the N-bit input signal, and uses the difference as a residual signal to encode a residual signal encoding processing unit. 17 to output.

  In step S16, the conversion table encoding processing unit 16 encodes the conversion table stored in the conversion table storage unit 11 for each divided area.

  In step S17, the residual signal encoding processing unit 17 encodes the residual signal of the difference between the signal after the bit depth conversion and inverse conversion and the input signal.

  In step S18, the encoded division information, the low bit depth signal of (N−Δ) bits, the residual signal, and the encoded data of the conversion table are combined and output as an encoded stream of the image encoding result.

  FIG. 3 is a diagram illustrating a configuration example of the bit depth conversion processing unit 10. As shown in FIG. 3, for example, the bit depth conversion processing unit 10 shown in FIG. 1 includes an area dividing unit 100, a class dividing unit 101, a conversion table creating unit 102, a pixel value converting unit 103, an error calculating unit 104, and a division determining unit. 105, a low bit depth signal generator 106 is provided. The conversion table storage unit 11 may be the same as that shown in FIG. 1 or may be prepared separately.

  The area dividing unit 100 performs a process of dividing the encoding target frame into M areas. There are a method of setting the division number M to a predetermined fixed value and a method of adaptively determining it as a variable value. First, an example in which the division number M is a fixed value will be described.

  FIG. 4 is a diagram illustrating an example of the region dividing method. For dividing the area, some division patterns may be determined in advance according to the division number M. For example, a method of repeating two divisions as shown in FIG. 4 may be used.

  In this example, as shown in FIG. 4A, division points are defined on the upper end and left end lines of the input image (frame), and the division points are sequentially divided. Here, an example in which there are six candidates for division points (division positions) from D1 to D6 is shown. As the number of division point candidates increases, optimization is promoted, but the calculation time increases.

  4 (B) to 4 (G) show examples in which the region is divided into two, (B) is divided at the dividing point D1, (C) is divided at the dividing point D2, ..., (G). Shows an example in the case of division at the division point D6.

  In the case of dividing into three or more divisions, the method of dividing one of the two divided regions into two by the same division method is repeated. 4 (H) to 4 (L) show an example in which the larger one of the regions divided into two as shown in FIG. 4 (E) is further divided into three and divided into three. Similarly, with respect to the division of four or more, any one of the three divided areas may be further divided into two.

If you want to repeat the division and select the area to be further divided,
(1) A method for selecting a region where the bit depth conversion error of the image in the region is large,
(2) A method for selecting a region having a large bit depth conversion error per unit area in the region,
Any method can be used.

  The class dividing unit 101 divides the range of pixel values represented by N bits into 2 (N−Δ) power classes using the Lloyd-Max algorithm for each divided area. The conversion table creation unit 102 creates a conversion table in which representative values of each class are arranged in order of class numbers from the division result of the class division unit 101 and stores the conversion table in the conversion table storage unit 11.

  The pixel value conversion unit 103 classifies the pixel value of the N-bit input signal into a class having the closest representative value in the conversion table, expresses the class number with (N−Δ) bits, and obtains a conversion result. The error calculation unit 104 calculates the sum of bit depth conversion errors for all the divided areas. The sum of bit depth conversion errors is the sum of squares for all pixel values of the difference between the pixel value of the input signal before conversion and the representative value of the converted class. The sum of absolute values may be used instead of the sum of squares.

  When the region is divided into two, for example, the processing from the region dividing unit 100 to the error calculating unit 104 is repeated for six types of division methods shown in FIGS. Even in the case of three or more divisions, region division is performed in the same manner by the above-described division method, and the sum of bit depth conversion errors is calculated.

  The division determination unit 105 determines the region division having the smallest value among the sums of bit depth conversion errors calculated by the error calculation unit 104 as the region division to be actually used, and determines the division point indicating the division position. Output information as split information. The low bit depth signal generation unit 106 converts the input signal of N bits into a low bit depth signal of (N−Δ) bits using the conversion table for each area created at the time of the adopted area division, Output.

  FIG. 5 is a flowchart of the bit depth conversion process performed by the bit depth conversion processing unit 10.

  In step S101, the area dividing unit 100 divides the area of the input image into M parts. In step S102, the class dividing unit 101 generates a histogram of pixel values for each divided area. In step S103, the centroid value of the generated histogram is obtained, and the histogram is divided into two using the obtained centroid value as the center value. In step S104, the smaller pixel value portion and the larger pixel value portion of the bisected histogram are divided into 2 (N−Δ−1) power classes at equal intervals. As a result, 2 (N−Δ) power classes are generated as a whole.

  In step S105, the centroid of the histogram is obtained in each class, and the obtained centroid is set as the representative point of the class. In step S106, the class is reset based on the interval between the representative points of each class, and according to the distance from each pixel value to the representative point, each pixel value is assigned to the class having the closest representative point. Recalculate the center of gravity. In step S107, it is determined whether or not the center of gravity has been updated by recalculating the center of gravity. If the center of gravity has been updated, steps S106 and S107 are repeated.

  In step S108, when the center of gravity is no longer updated, the conversion table creation unit 102 creates a conversion table from the representative point values (representative values) of the class at that time. A conversion table is created for each divided area.

  In step S109, the pixel value conversion unit 103 converts the pixel value (N bits) in each class into a class number of (N−Δ) bits for each region.

  In step S110, the error calculation unit 104 calculates the error between the input image and the converted image, that is, the sum of the bit depth conversion errors for all regions, and stores the error together with the division information.

  In step S111, it is determined whether or not the error of bit depth conversion has been calculated for all the region division candidates. If not completed, the process returns to step S101, and the region is similarly divided for other division points. The process is repeated until the calculation of the error for the candidate for region division is completed.

  In step S112, when the calculation of errors for all region division candidates is completed, the division determination unit 105 selects a region division that minimizes the error, and outputs division information indicating the division position. In addition, the conversion table for each region adopted at the time of the region division and the (N−Δ) bit signal after bit depth conversion by the low bit depth signal generation unit 106 are output.

  FIG. 6 is a flowchart of the bit depth conversion process when the area division number M is variable and is determined adaptively. In the example of FIG. 5, it is assumed that the division number M of the area is determined in advance. However, as shown in FIG. 6, processing that adaptively determines the optimal division number is also possible.

  First, in step S120, the initial value of the division number M is set to M = 2, and low bit depth signal conversion processing is performed for M regions. Here, the processing of steps S101 to S111 described in FIG. 5 is executed.

  Next, in step S121, the division determination unit 105 shown in FIG. 3 estimates the sum of the code amount of the conversion table and the code amount of the residual signal in the M optimal region divisions for the entire region of the screen.

  In step S122, it is determined whether to change the number of divisions. Whether or not to change the number of divisions may be determined whether or not the processing up to the upper limit value of the predetermined number of divisions has been completed. For example, the code amount of the conversion table and the residual signal The determination may be made based on whether the sum with the code amount is greater than a certain threshold value compared to the previous division number.

  When changing the division number, the process proceeds to step S123, 1 is added to the division number M, the process returns to step S120, and the process is repeated in the same manner.

  Thereafter, in step S124, the division determination unit 105 determines the number of divisions that minimizes the sum of the code amount of the conversion table and the code amount of the residual signal. In step S125, the division information indicating the determined number of divisions and the optimum division position at the number of divisions, the conversion table for each region, and the signal after the low bit depth conversion are output.

  FIG. 7 is a diagram illustrating a configuration example of the conversion table encoding processing unit 16. FIG. 8 is a flowchart of conversion table encoding processing by the conversion table encoding processing unit 16.

  The conversion table encoding processing unit 16 includes a fixed-length encoding processing unit 161, an intra-table difference calculation unit 162, a Golomb encoding processing unit 163 that performs Golomb encoding of the intra-table difference, an inter-table difference calculation unit 164, A Golomb encoding processing unit 165 and a code length determination unit 166 for Golomb encoding the inter-table difference are provided.

  The operation of the conversion table encoding processing unit 16 will be described with reference to the flowchart of FIG. First, in step S130, the fixed-length encoding processing unit 161 performs fixed-length encoding of each element of the conversion table with N bits.

  In step S131, the in-table difference calculation unit 162 calculates a difference in the conversion table, that is, a difference from the previous element for each element of the conversion table, and the Golomb encoding processing unit 163 calculates the difference information to Golomb. Encode.

  In step S132, the inter-table difference calculation unit 164 calculates the difference between the conversion tables, that is, the difference between each element of the other encoded conversion table and each element of the conversion table to be encoded. The difference information is Golomb encoded by the Golomb encoding processing unit 165. The other encoded conversion table may be the previous conversion table or the conversion table at the position where the area one frame before most overlaps.

  In step S133, the code length determination unit 166 selects the encoded data of the fixed length encoding, the difference in the table, and the difference between the tables that is the output of steps S130 to S132, and selects which code length is the smallest. Along with information indicating whether the encoding method is used, the conversion table encoded data having the minimum code length is output.

[Image decoding device]
FIG. 9 is a diagram showing a configuration example of an image decoding apparatus related to the present invention.

  The image decoding apparatus 2 includes a stream decomposition unit 20, a division information decoding processing unit 21, a conversion table decoding processing unit 22, a conversion table storage unit 23, a low bit depth signal decoding processing unit 24, and a residual signal decoding process. A unit 25, an inverse bit depth conversion processing unit 26, and an adder 27 are provided to decode an encoded stream that is an output of the image encoding device 1 shown in FIG.

  FIG. 10 shows a flowchart of the decoding process executed by the image decoding device 2. First, in step S20, the stream decomposing unit 20 inputs an encoded stream to be decoded, and the encoded stream is divided into divided information encoded data, conversion table encoded data, low bit depth signal encoded data, Decompose into residual signal encoded data.

  In step S21, the division information decoding processing unit 21 decodes division information indicating how the frame area is divided. In addition, when the division information is encoded together with other encoding parameters or the like, decoding is performed in accordance with the decoding of the encoding parameters or the like.

  In step S22, the conversion table decoding processing unit 22 decodes the encoded data of the conversion table created for each divided area based on the division information decoded by the division information decoding processing unit 21, and stores the conversion table. Stored in the unit 23.

  In step S23, the residual signal decoding processing unit 25 decodes the residual signal encoded data, and in step S24, the low bit depth signal decoding processing unit 24 performs (N−Δ) -bit low bit depth signal encoding. Decrypt the data. In this decoding, for example, conventional H.264. A decoding process similar to that of a decoder for H.264 encoded data is performed.

  In step S25, the inverse bit depth conversion processing unit 26 converts the decoded low bit depth signal into an N-bit high bit depth signal using the conversion table for each area stored in the conversion table storage unit 23. . That is, the representative value of the class corresponding to the class number indicated by the low bit depth signal is set as the pixel value of each pixel.

  In step S26, the adder 27 adds the decoded signal of the residual signal and the high bit depth signal converted by the inverse bit depth conversion processing unit 26, and synthesizes and outputs the decoded signal of N bits.

  FIG. 11 is a diagram illustrating a configuration example of the conversion table decoding processing unit 22. The conversion table decoding processing unit 22 includes an encoding method determination unit 221, a fixed length decoding processing unit 222, an in-table difference decoding processing unit 223, and an inter-table difference decoding processing unit 224.

  FIG. 12 is a flowchart of conversion table decoding processing performed by the conversion table decoding processing unit 22. In step S30, the encoding method determination unit 221 uses the encoding method information included in the conversion table encoded data to encode the conversion table using fixed length encoding, intra-table difference, or inter-table difference. Determine.

  In step S31, based on the determination result of the encoding method, if the fixed-length encoding is used, the fixed-length decoding processing unit 222 decodes the conversion table. The difference information is Golomb decoded, and if it is an inter-table difference, the inter-table difference decoding processing unit 224 performs Golomb decoding of the inter-table difference information.

[Effect verification experiment]
In order to verify the effect of the present invention, the following experiment was conducted. The image used as the encoding target is a 10-bit 4: 2: 0 YCbCr image (7 types), and the resolution is 1920 × 1080. In the bit depth conversion, 10 bits were converted to 8 bits. That is, in the above embodiment, N = 10 and Δ = 2.

  As the low bit depth signal encoding processing unit 12, the decoding processing unit 13, and the residual signal encoding processing unit 17 shown in FIG. H.264 / AVC reference software (JM 14.0) was used. In the case where the region division is not performed, the low bit depth signal encoding processing unit 12, the decoding processing unit 13, and the residual signal encoding processing unit 17 are respectively connected to the conventional encoding processing unit 301, FIG. This corresponds to the decoding processing unit 302 and the encoder 305.

In the experiment, the following four types of experiments were performed for seven types of image sequences (video 1 to video 7). Each video 1 to video 7 has 60 frames.
[Experiment 1]: [Prior Art] The conversion table was encoded at a fixed length without performing region division (encoding by the image encoding device in FIG. 14).
[Experiment 2]: The number of area divisions was set to a fixed value of 2, and the conversion table was encoded with a fixed length for each area.
[Experiment 3]: The region division number was variable, and the optimum division number was obtained to perform region division. For each region, the conversion table was encoded with a fixed length.
[Experiment 4]: The area division number is variable and the optimum division number is obtained to perform area division. For each area, the conversion table has the shortest code length among the fixed length, intra-table difference, and inter-table difference. Adaptive coding using coding scheme was performed.

  In the above experiment, a 10-bit input sequence is converted to H.264. It was investigated how much the code amount was reduced when the image quality was comparable to the result of direct encoding with H.264 / AVC reference software (JM 14.0). That is, the bit rates were compared when the image quality (PSNR) was the same.

  The code amount reduction rate of the experimental result was as shown in FIG. FIG. 13B shows the numerical values shown in FIG. 13A in a graph. Compared to Experiment 1, Experiments 2 to 4 confirmed that the reduction rate of the code amount was larger for any of the videos 1 to 7. In particular, in Experiment 4, it was found that the code amount can be reduced by about 11% on average as compared with the case of direct encoding using the reference software JM.

  In Experiments 3 and 4, when the number of divisions is variable and the area is divided by the optimum number of divisions, the number of divisions of each of video 1 to video 7 is as shown in FIG. From these experimental results, it was confirmed that the optimal number of segmentation varies greatly depending on the local properties of the image.

  The above image encoding and decoding processes can be realized by a computer and a software program, and the program can be recorded on a computer-readable recording medium or provided through a network.

DESCRIPTION OF SYMBOLS 1 Image coding apparatus 10 Bit depth conversion process part 11 Conversion table memory | storage part 12 Low bit depth signal encoding process part 13 Decoding process part 14 Inverse bit depth conversion process part 15 Subtractor 16 Conversion table encoding process part 17 Residual signal Encoding processing unit 18 Division information encoding processing unit 100 Region division unit

Claims (5)

An image signal is input, bit depth conversion processing into a low bit depth signal is performed, the low bit depth signal is encoded, a residual signal between the input image signal and the low bit depth signal is encoded, and the bit In the image encoding device that encodes the conversion table used for the depth conversion process and outputs the encoded stream,
Area dividing means for dividing the screen of the input image signal into a plurality of areas;
Bit depth conversion processing means for converting an input image signal into a low bit depth signal using a conversion table created for each of the divided areas;
Low bit depth signal encoding means for encoding the low bit depth signal;
Decoding means for decoding the encoded low bit depth signal;
An inverse bit depth conversion processing means for converting the decoded low bit depth signal into the bit depth of the input image signal using the conversion table;
Residual signal generation means for calculating a difference between the output signal of the inverse bit depth conversion processing means and the input image signal;
Residual signal encoding processing means for encoding the residual signal;
Conversion table encoding processing means for encoding the conversion table;
Division information encoding means for encoding division information indicating positions divided by the area dividing means;
Means for outputting an encoded stream including encoded data output from the low bit depth signal encoding means, the residual signal encoding processing means, the conversion table encoding processing means, and the division information encoding means. ,
The region dividing means sequentially performs region division for each of a plurality of determined division positions, and performs region division that minimizes an error over the entire screen area between the input image signal and the image signal after the bit depth conversion processing. , And finally dividing into regions to be used for encoding . Further, when the region is divided into three or more regions in the region division, the region is divided into two, and the input image signal is divided into the divided regions. And an image signal after the bit depth conversion process , an image coding apparatus characterized by repeating the process of further dividing the region having the larger error . The image encoding device according to claim 1,
The number of areas divided by the area dividing means is variable;
An image encoding apparatus comprising: means for determining a division number that minimizes the sum of the code amount of the conversion table and the code amount of the residual signal over the entire area of the screen as the division number of the area. The image encoding device according to claim 1 or 2 ,
The conversion table encoding processing means includes:
Encoding means for encoding the elements of the conversion table at a fixed length;
Encoding means for encoding an element of the conversion table with a difference from the previous element in the same conversion table;
Among the encoding means for encoding the element of the conversion table with the difference from the element in the other encoded conversion table,
At least one of a plurality of encoding means,
An image encoding apparatus, wherein the conversion table is encoded by selecting an encoding unit having a minimum code amount from the plurality of encoding units. An image signal is input, bit depth conversion processing into a low bit depth signal is performed, the low bit depth signal is encoded, a residual signal between the input image signal and the low bit depth signal is encoded, and the bit In the image encoding method for encoding the conversion table used for the depth conversion process and outputting the encoded stream,
An area dividing step for dividing an input image signal screen into a plurality of areas;
A bit depth conversion processing step for converting an input image signal into a low bit depth signal using a conversion table created for each of the divided areas;
A low bit depth signal encoding step for encoding the low bit depth signal;
A decoding step of decoding the encoded low bit depth signal;
An inverse bit depth conversion processing step for converting the decoded low bit depth signal into the bit depth of the input image signal using the conversion table;
A residual signal generation step of calculating a difference between the output signal of the inverse bit depth conversion processing step and the input image signal;
A residual signal encoding processing step for encoding the residual signal;
A conversion table encoding processing step for encoding the conversion table;
A division information encoding step for encoding division information indicating the position divided in the region division step;
Outputting a coded stream including the coded data output in the low bit depth signal coding step, the residual signal coding processing step, the conversion table coding processing step, and the division information coding step. Have
In the region dividing step, region division is sequentially performed for each of a plurality of determined division positions, and region division that minimizes an error over the entire screen area between the input image signal and the image signal after the bit depth conversion processing is performed. , And finally dividing into regions to be used for encoding . Further, when the region is divided into three or more regions in the region division, the region is divided into two, and the input image signal is divided into the divided regions. And an image signal after the bit depth conversion processing is repeated to further divide the region having the larger error into two . The image coding program for functioning a computer as said each means with which the image coding apparatus described in any one of Claim 1- Claim 3 is provided.

Images