JP2015088845A - Data compression device and data compression method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a data compression method of which the compressibility is high regardless of a magnitude of an amplification factor of data and which employs encoding capable of prompt processing, and a data compression device.SOLUTION: RAW image data obtained by an image sensor 11 are amplified by a digital gain unit 16 while using a magnification that is set in a gain setting unit 21. The amplified RAW image data are coded and compressed by a Huffman coding unit 17 and stored in a memory (SDRAM) 18. In the Huffman coding unit 17, if the magnification that is set by the gain setting unit 21 is less than double, coding is performed while using a standard Huffman table and if equal to or more than double, coding is performed while using a correction Huffman table. In the correction Huffman table, a Huffman code of a larger bit number is allocated to a category where a frequency of appearance is made 0 by amplification.

Description

  The present invention relates to a data compression method using encoding.

  For example, in a digital camera, the ISO sensitivity is artificially changed by adjusting the gain of RAW image data obtained by the image sensor. As the number of pixels increases, a configuration is also known in which image data is compressed irreversibly and using a variable-length encoding method when RAW image data is temporarily stored in an SDRAM. However, there is a problem that an increase in gain or data compression by irreversible and variable-length coding deteriorates the S / N ratio. To solve such a problem, a configuration has been proposed in which deterioration of image quality is prevented by not compressing RAW image data when a set gain is equal to or greater than a threshold value (see Patent Document 1).

JP 2008-1113070 A

  On the other hand, a method of reversibly compressing RAW image data using a Huffman code is also known. In data compression using the Huffman code, it is necessary to assign the Huffman code in accordance with the appearance frequency of the data. That is, in order to increase the compression rate by optimal encoding, it is necessary to create a data histogram, create a table to which Huffman codes are assigned based on the appearance frequency of data, and then encode the data. However, since RAW image data compression requires a quick process, the process cannot be applied as it is.

  Therefore, when Huffman compression is used for compression of RAW image data, the appearance frequency of data obtained when a typical subject image is taken with standard ISO sensitivity is simulated in advance, and a standard Huffman table is set to 1. And RAW image data is encoded using the table. However, in the variable length encoding process using a standard Huffman table, there is a problem that when the gain (amplification factor) increases, the compression efficiency is greatly impaired, and in some cases, the capacity becomes larger than the original data.

  An object of the present invention is to provide a compression method and a data compression apparatus using encoding that can perform a quick process with a high compression ratio regardless of the magnitude of the data amplification factor.

  The data compression apparatus according to the present invention compresses input digital data based on a code table, an amplification unit that amplifies input digital data at a predetermined magnification, gain setting means that sets a value of the predetermined magnification, and digital data amplified by the amplification unit. A compression unit and code table switching means for switching the code table to either the standard code table or the correction code table according to the set magnification value, and the standard code table is used when the magnification is 1, It is characterized in that the correction code table is used when is 2 or more.

  The digital data is preferably digital image data, and the code table is divided into a plurality of categories in which the difference values of the digital image data are classified according to their sizes. A code is assigned to each category, and the difference between the difference values included in each category is represented by additional bits.

  In the correction code table, a code having a larger number of bits is assigned to a category whose appearance frequency decreases due to amplification. The standard code table includes a first category group including a category including a difference value equal to or less than a predetermined value, and a second category group including a category including a difference value larger than the predetermined value. A category including a value is assigned a smaller code, and in the second category group, a category including a larger difference value is assigned a larger code.

In the correction code table, the code assigned to the second category group in the standard code table is assigned to a category whose appearance frequency is 0 by amplification. As a code assigned to a category whose appearance frequency decreases due to amplification, it is preferable to select a code having a larger number of bits from the second category table of the standard code table. The categories are classified into categories 0 to N as the difference value increases. When n is an integer from 0 to N, the difference values classified into the category n are (−2 ⁿ +1) to (−2). ^n-1 ) and (2 ^n-1 ) to (2 ⁿ -1), and when m is an integer of 1 or more, the correction code table is switched every time the magnification increases by 2 ^m . The image data is, for example, RAW image data.

  A digital camera according to the present invention includes any one of the data compression apparatuses described above.

  The compression method using encoding according to the present invention includes a step of amplifying input digital image data at a predetermined magnification, a step of setting a value of the predetermined magnification, and compressing the amplified digital image data based on a code table. And a step of switching the code table to either the standard code table or the correction code table in accordance with the set magnification value.

  According to the present invention, it is possible to provide a compression method and a data compression apparatus using encoding that can perform high-speed processing and a high compression rate regardless of the magnitude of the data amplification rate.

It is a block diagram which shows the flow of the image data in the image data compression apparatus of this embodiment. It is a table | surface which shows the relationship between the category used for the Huffman compression of this embodiment, and the difference value of image data. It is a standard Huffman table used for Huffman compression of this embodiment. It is the correction Huffman table of this embodiment used when magnification is 2 times or more and less than 4 times. It is the correction Huffman table of this embodiment used when a magnification is 4 times or more and less than 8 times. It is the correction Huffman table of this embodiment used when a magnification is 32 times or more. A compression rate when encoding is performed using only the standard Huffman table and when encoding is performed using the modified Huffman table is shown. It is a flowchart which shows the process sequence of the Huffman compression of this embodiment.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a flow of image data in an image data compression apparatus according to an embodiment of the present invention.

  In this embodiment, the image data compression apparatus is mounted on the digital camera 10, for example. In the digital camera 10, an image signal obtained by the image sensor 11 is converted into digital image data via an image sensor I / F (interface) 12 and input to the digital signal processing circuit 13. The digital signal processing circuit 13 includes, for example, a black level correction unit 14, an AE detection unit 15, a digital gain unit 16, and a Huffman coding unit 17, and is connected to a memory 18 such as an SDRAM. The memory 18 is connected to an interface (SD card I / F) 19 corresponding to a card memory such as an SD card, and can output and store image data to a memory card (SD card) 20. The image data can be output from the memory 18 to a monitor device (not shown).

  The digital image data input to the digital signal processing circuit 13 is first subjected to black level correction by subtracting the offset of optical black (OB) data in the black level correction unit 14. The digital image data subjected to the black level correction is output to the AE detection unit 15 and the digital gain unit (amplifier circuit) 16. In the AE detection unit 15, white balance data is detected and held in a memory (SDRAM) 18. Further, the digital gain unit 16 amplifies the digital image data at a set predetermined magnification, and the amplified digital image data is temporarily held in the memory (SDRAM) 18 as RAW image data.

  Note that the magnification (amplification factor) in the digital gain unit 16 corresponds to the ISO sensitivity and is set in the gain setting unit 21. In the gain setting unit 21, the ISO sensitivity (magnification) is manually set by the user in the manual mode, and is automatically set in the auto mode based on the result of automatic photometry.

  The digital image data (RAW data) temporarily held in the memory (SDRAM) 18 via the digital gain unit 16 is sent to the Huffman encoding unit 17 and subjected to Huffman compression processing, which will be described later, and the memory ( (SDRAM) 18 is stored as compressed RAW data, and is stored in the SD card 20 via the SD card I / F 19 as necessary.

  Next, the principle of the image data compression method (Huffman compression processing) using the Huffman code of this embodiment will be described with reference to FIGS.

  FIG. 2 shows the relationship between image data difference values and categories in the Huffman compression processing of this embodiment. In FIG. 2, the difference value is a difference in image data (pixel value) between adjacent pixels (hereinafter referred to as adjacent pixels) in the same color, and for example, the left adjacent pixel is used as the adjacent pixel (usually other colors) Adjacent pixels of the same color). When the left adjacent pixel is used as the adjacent pixel, the difference value of the leftmost pixel is a difference value from a predetermined value given in advance, a difference value from the pixel value of the upper pixel, or the previous line. A difference value from the pixel value on the right end is used, but is not limited thereto.

As shown in FIG. 2, in this embodiment, category 0 has a difference value of 0, category 1 has a difference value of {-1, 1}, category 2 has a difference value of {-3, -2. , 2, 3}, category 3 has a difference value of {−7, −6, −5, −4, 4, 5, 6, 7}, and the larger the difference value, the higher the number. The In this embodiment, the difference value having the largest difference, that is, an integer of {−16383 to −8192, 8192 to 16383}, is assigned to the category 14. That is, the difference value is categorized n ^{is, {(- 2 n +1)} ~ (-2 n-1), (2 n-1) ~ (2 n -1)} is an integer (here N is an integer of 0 to 14).

  FIG. 3 is a standard Huffman table of the present embodiment set based on the category of FIG. Here, the standard Huffman table is used when the magnification (amplification factor) set in the gain setting unit 21 is less than 2, assuming a typical subject image and amplifying it with a magnification of 1 (that is, not amplifying). The Huffman code is assigned to each category in accordance with the appearance frequency of the difference value.

  In the standard Huffman table of FIG. 3, for reasons described later, an 8-bit Huffman code is assigned to category 0, and then a 7- to 3-bit Huffman code is sequentially assigned to categories 1 to 5, respectively. In categories 5 to 7, a 3-bit Huffman code is assigned, and in categories 8 and 9, a 2-bit Huffman code that is the minimum number of bits is assigned. Thereafter, 9 to 11-bit Huffman codes are assigned to categories 10 to 12, and 12-bit Huffman codes that are the maximum number of bits are assigned to categories 13 and 14.

The additional bits in FIG. 3 are bits for identifying each of the difference values included in each category. For example, since category 0 is only when the difference value is 0, no additional bit is required and no additional bit is assigned. Since there are two kinds of difference values included in category 1, {-1, 1}, the number of bits necessary and sufficient to express the difference between the two is 1, and 1 bit is allocated as an additional bit. Further, there are four types of difference values {-3, -2, 2, 3} included in category 2, and the number of additional bits necessary and sufficient for identification is 2 bits. In this embodiment, the category n includes 2 · (2 ⁿ −2 ⁿ⁻¹ ) = 2 ⁿ difference values, and the number of additional bits necessary and sufficient to represent these is n bits.

  That is, when the image data is encoded using the Huffman table, the total number of bits after encoding of the image data included in each category is the number of bits of the Huffman code assigned to the category and the additional bits of the category. The sum of numbers. In FIG. 3, the total number of bits when the image data of each category is encoded is shown in the rightmost column.

  In encoding using the Huffman table, data with a high appearance frequency is converted into a code with a smaller total number of bits, and data with a low appearance frequency is converted into a code with a larger total number of bits, thereby increasing the data compression rate. be able to. However, the frequency of appearance of the difference value in any photographed image cannot be grasped unless a histogram of difference values is created after photographing. Since temporary processing of RAW image data requires high-speed processing, it is not realistic to create a histogram of difference values each time an image is taken.

  Therefore, when Huffman compression is used for RAW image data in a digital camera, the appearance frequency of each category with respect to a typical subject image is calculated in advance, a standard Huffman table is created, and the RAW image is generated using the Huffman table. Huffman compression of image data is performed.

  In the example of FIG. 3, the appearance frequency of the difference value gradually increases from the category 0, and in the category 9, the appearance frequency of the difference value becomes substantially maximum. And in category 10, the appearance frequency decreases rapidly, and from category 10 to category 14, the appearance frequency of the difference value gradually decreases. When the difference values are classified into categories by the method as shown in FIG. 2, the appearance frequency of the difference values of the image data generally increases gradually as the category number increases from 0, and is almost the maximum value in a specific category. (First category group). In the category immediately after that, the appearance frequency decreases rapidly, and thereafter, the appearance frequency tends to gradually decrease (second category group) as the category number increases. In the standard Huffman table of FIG. 3, categories 0 to 9 are classified as first category groups, and categories 10 to 14 are classified as second category groups.

  Therefore, in the standard Huffman table of FIG. 3, the total number of bits of a category having a high appearance frequency is calculated from the relationship between the number of additional bits of each category and the difference value appearance frequency of each category in the image data of the gain magnification 1 of the typical subject image. A Huffman code is assigned to each category so as to be as small as possible, and the compression rate of the RAW image data is increased. In general, since the appearance frequency of each category included in the second category group is lower than the appearance frequency of each category included in the first category group, in the standard Huffman table, the second category group is compared with the first category group. A Huffman code having a larger number of bits is allocated. Therefore, in the standard Huffman table, the Huffman code having the maximum number of bits is assigned to the category having the largest category number.

However, when the image data is amplified, for example, twice in the digital gain unit 16, all the values of the digital image data subjected to gain processing are all doubled, so that all the difference values are 2 or more or -2 or less. The difference value of ± 1 included in category 1 does not exist in principle. Similarly, when the magnification is 4 or more, the appearance frequency of the difference values {± 1, ± 2, ± 3} included in categories 1 and 2 is zero. Generally, when the magnification is 2 ^m or more (m: an integer of 1 or more), the appearance frequency of the difference values included in category 1 to category m is 0.

  Therefore, the number of bits of the Huffman code assigned to the second category group using the standard Huffman table when the magnification is 1 even though the ISO sensitivity (magnification) is changed as in the conventional digital camera. Is larger than the number of bits of the Huffman code assigned to the category included in the first category group whose appearance frequency becomes 0 in principle due to the change in ISO sensitivity (magnification), and efficient compression is performed. I will not.

  From the above, in this embodiment, when the magnification set by the ISO sensitivity is 2 or more, encoding is performed using a modified Huffman table in which the allocation of Huffman codes to each category is changed with respect to the standard Huffman table. As a result, the compression rate of the image data is increased. That is, the Huffman code assigned to the category whose difference value appearance frequency is 0 by amplification is replaced with one of the Huffman codes assigned to the second category group in the standard Huffman table, and the Huffman code is replaced. The Huffman codes assigned to the categories after the above category are sequentially assigned by shifting to the larger category number by the amount of replacement. At this time, the specific category corresponding to the boundary between the first category group and the second category group is also shifted to the larger category number as the Huffman code is shifted.

In particular, in the present embodiment, every time the magnification is increased by 2 ^m , the maximum Huffman code in the second category group is assigned to the category of the first category group in which the appearance frequency is zero. 4 shows that when the set gain (magnification) is 2 times or more and less than 4 times, FIG. 5 shows when the set gain is 4 times or more and less than 8 times, and FIG. 6 shows when the set gain is 32 times or more. A modified Huffman table of the present embodiment is shown. Note that the modified Huffman table in the case where the set gain is 8 times or more to less than 16 times or 16 times or more to less than 32 times is omitted, but is defined by the same method.

  When the magnification is 2 times or more and less than 4 times, the appearance frequency of category 1 is 0 as described above. Therefore, in the modified Huffman table of FIG. 4, the Huffman assigned to category 14 in the standard Huffman table of FIG. Codes 111111111111 are assigned to category 1, and Huffman codes assigned to category 1 to category 13 in the standard Huffman table are assigned to category 2 to category 14, respectively. In FIG. 4, the first category group is category 0 to category 10, and the second category group is category 11 to category 14.

  When the magnification is 4 times or more and less than 8 times, as described above, the appearance frequency of category 2 becomes 0 in addition to category 1 in FIG. 4, so in the modified Huffman table in FIG. Huffman codes 111111111110 are further assigned as category 2 Huffman codes, and the Huffman codes assigned to categories 1 to 12 in the standard Huffman table are assigned to categories 3 to 14, respectively. In FIG. 5, the first category group is category 0 to category 11, and the second category group is category 12 to category 14.

  When the magnification is 32 times or more, the appearance frequency of category 1 to category 5 is 0, and category 1 to category 5 are assigned to category 14 to category 12 of the second category group in the standard Huffman table of FIG. Huffman codes 111111111111, 111111111110, 11111111110, 11111111110, and 111111110 that have been assigned are assigned to categories 1 to 5, respectively, and Huffman codes assigned to categories 1 to 9 in the standard Huffman table are assigned to categories 6 to 14 respectively. Assigned. At this time, all of category 0 to category 14 belong to the first category group.

That is, when the Huffman code of category n in the standard Huffman table is represented by C (n), in the modified Huffman table corresponding to a magnification of 2 ^m , i is an integer equal to or greater than 1, and the category (m + i) has Huffman The code is assigned C (i). If the maximum value of the category number is N, the Huffman code of category N in the modified Huffman table corresponding to the magnification of 2 ^m is C (N−m). In the present embodiment, the category 0 Huffman code is the same in both the standard Huffman table and the modified Huffman table. The Huffman codes assigned to categories 1 to m in the modified Huffman table are C (N) to C (N−m−1) in this embodiment, but the appearance frequencies of categories 1 to m are all 0. So these assignments are out of order.

  FIG. 7 shows the compression rate when encoding is performed using only the standard Huffman table (when the table is fixed), and compression when encoding is performed using the modified Huffman table of the present embodiment (when switching the table). The rate is indicated. In FIG. 7, when the magnification (amplification factor) of the image data is 1, 2, 4, 8, 16, and 32 times, the magnification when the table is fixed and the magnification when the table is switched are shown. Here, the magnification is defined by (data amount after compression) / (original data amount). As shown in FIG. 7, it can be seen that at a magnification of 2 or more, the compression rate is higher when the modified Huffman table is used than when the standard Huffman table is used.

  FIG. 8 is a flowchart of the Huffman compression processing of this embodiment. The Huffman compression processing of this embodiment will be described with reference to FIGS. This process is executed by the Huffman encoding unit 17 when, for example, a release operation is performed and the RAW image data is temporarily stored in the memory (SDRAM) 18 of FIG.

  In step S100, the gain (magnification) set in the gain setting unit 21 is acquired based on the ISO sensitivity. In step S102, a Huffman table used for compression processing (encoding) is selected based on the obtained magnification. . That is, in this embodiment, when the gain (magnification) is less than 2 times, the standard Huffman table of FIG. 3 is selected, and when it is 2 times or more and less than 4 times, the modified Huffman table of FIG. When the correction Huffman table of FIG. 5 is 8 times or more and less than 16 times, or 16 times or more and less than 32 times, the correction Huffman table suitable for each magnification is selected. When it is 32 times or more, the modified Huffman table shown in FIG. 6 is selected.

  In step S104, pixel difference values in the RAW image data temporarily stored in the memory (SDRAM) 18 are sequentially calculated via the digital gain unit 16, and which of the Huffman tables in which the difference value is selected in step S106. Whether it is included in the category is determined. In step S108, the RAW image data of the same pixel is encoded based on the selected Huffman table and sequentially stored in the memory (SDRAM) 18. In step S110, it is determined whether the processing in steps S104 to S108 has been performed on the RAW image data of all the pixels. If not completed, the processing returns to step S104, and the same processing is repeated for the next pixel. The encoded data is written to the next address in the previous data. On the other hand, if it is determined in step S110 that the processes in steps S104 to S108 have been completed for all pixels, the Huffman compression process is terminated. This process is performed for each color data.

  As described above, according to the present embodiment, even when the amplified image data is Huffman-compressed, the image data can be quickly compressed at a high compression rate using the modified Huffman table regardless of the amplification factor (magnification). Therefore, the Huffman compression can be used effectively to compress the RAW image data.

  Although a modified Huffman table may be prepared in advance for each magnification, in this embodiment, a modified Huffman table that can be adapted to each magnification can be created very easily. Can be produced each time according to the selected magnification.

  The present invention is effective in applications where data amplified at a plurality of magnifications needs to be compressed quickly, reversibly, and efficiently, and is not limited to digital camera applications. RAW image data However, the present invention is not limited to compression of image data, and further compression of image data. Further, the digital camera of the present embodiment may be mounted on a mobile phone, a smart phone, or the like.

  The standard Huffman table is not limited to this embodiment, and is determined according to an assumed typical image. In addition, a plurality of standard Huffman tables may be prepared under shooting conditions (shooting modes) other than ISO sensitivity (magnification). In this case, the present embodiment is applied to each standard Huffman table according to the magnification. Similar processing is performed. Note that, as the shooting mode, for example, a mode corresponding to the subject and the shooting environment such as a mode for shooting a landscape, a mode for shooting a person, and a mode for shooting a night view can be considered.

  In this embodiment, the category is divided into 15 categories of category 0 to category 14, but the number of categories is not limited to this, and the number of modified Huffman tables is increased or decreased according to the amplification factor and the number of categories. . Note that a configuration in which Huffman compression is not performed when the magnification exceeds a predetermined value may be employed.

DESCRIPTION OF SYMBOLS 10 Digital camera 11 Image sensor 13 Digital signal processing circuit 16 Digital gain unit 17 Huffman encoding unit 18 Memory 21 Gain setting unit

Claims (10)

An amplification unit for amplifying input digital data at a predetermined magnification;
Gain setting means for setting a value of the predetermined magnification;
A compression unit for compressing the digital data amplified by the amplification unit based on a code table;
A code table switching means for switching the code table to either a standard code table or a correction code table according to the set magnification value;
The data compression apparatus, wherein the standard code table is used when the magnification is 1, and a correction code table is used when the magnification is 2 or more.   The digital data is digital image data, and the code table is divided into a plurality of categories in which the difference values of the digital image data are classified according to their sizes, and a code is assigned to each of the categories and each category is assigned. 2. The data compression apparatus according to claim 1, wherein the difference between the difference values included is represented by additional bits.   3. The data compression apparatus according to claim 1, wherein a code having a larger number of bits is assigned to a category whose appearance frequency has decreased due to the amplification in the correction code table.   The standard code table includes a first category group including a category including a difference value equal to or less than a predetermined value, and a second category group including a category including a difference value larger than the predetermined value. In the first category group, The category including a larger difference value is assigned a smaller code, and the category including a larger difference value is assigned a larger code in the second category group. The data compression apparatus described.   5. The data compression apparatus according to claim 4, wherein the correction code table assigns a code assigned to the second category group in the standard code table to a category whose appearance frequency is 0 by the amplification.   6. The code having a larger number of bits is selected from the second category table of the standard code table as a code assigned to a category whose appearance frequency is 0 by the amplification. Data compression device. The category is classified into category 0 to category N as the difference value increases. When n is an integer from 0 to N, the difference value classified into category n is (−2 ⁿ +1) to (−2). -2 ^n-1 ) and (2 ^n-1 ) to (2 ⁿ -1), where m is an integer of 1 or more, the correction code table is switched every time the magnification is increased by 2 ^m. The data compression apparatus according to claim 2, wherein the data compression apparatus is a data compression apparatus.   The data compression apparatus according to any one of claims 2 to 8, wherein the image data is RAW image data.   A digital camera equipped with the data compression device according to claim 1. Amplifying input digital image data at a predetermined magnification;
Setting a value of the predetermined magnification;
Compressing the amplified digital image data based on a code table;
And a step of switching the code table to either a standard code table or a modified code table in accordance with the set value of the magnification.

Images