JP4344391B2 - Image processing device - Google Patents

Description

  The present invention relates to an image processing apparatus, and more particularly to an image processing apparatus compliant with JPEG-2000.

  In recent years, “JPEG-2000” has been standardized as a next-generation international standard for still image compression technology.

  This "JPEG-2000" is a compression method succeeding "JPEG (Joint Photo-graphic Experts Group)", which is currently the mainstream still image compression technology for the Internet and digital still cameras. However, the part called “Part-1” that defines data compression technology is an international standard.

  In “JPEG-2000”, wavelet transform is adopted, which is very different from the compression technique used in the current “JPEG”. A general operation of encoding / decoding by “JPEG-2000” using wavelet transform will be described with reference to FIG.

  2 is configured to encode the image data input from the left side in the figure and output from the right side in the figure, and to decode the code data input from the right side in the figure and output from the left side in the figure It is.

More specifically, the color conversion unit 10 performs color conversion on the input image data. For example, image data input in the RGB (R = Red, G = Green, B = Blue) format is converted into a YCbCr (Y = luminance, Cb, CR = color difference) format, or vice versa. Conversion is performed. As an example, an equation (Equation 1) for color conversion of image data in RGB format is shown below.
G = Yr− (Cr + Cb−2) / 4 Y = (R + 2G + B) / 4
R = Cr + G Cr = RG
B = Cb + G Cb = BG (Formula 1)
However, the color conversion unit 10 is not necessarily a necessary configuration and functions as necessary.

  Next, in the example illustrated in FIG. 2, the wavelet transform unit 20 performs a plurality of levels of two-dimensional discrete wavelet transform (DWT) on each component of the image data independently. At this time, it is possible to divide the image to be processed into a plurality of rectangular blocks called tiles and perform two-dimensional discrete wavelet transform. However, when processing is performed by dividing an image into a plurality of tiles, the wavelet transform unit 20, the quantization unit 30, and the entropy coding unit 60 perform each process in tile units.

When wavelet transform is applied to a component that has not been subsampled Below, image data composed of three components that are not subsampled (for example, component 0 = Y, component 1 = Cb, and component 2 = Cr) may be 128 ×. An example will be described in which division into tiles having a size of 128 pixels is performed and level 3 two-dimensional discrete wavelet transform is performed.

An equation (equation 2) of an integer discrete wavelet transform (DWT) capable of reversible transformation, which is one of the wavelet transformations adopted in “JPEG-2000”, is shown.
L (k) = x (2k) + (H (k) + H (k + 1)) / 4
H (k) = x (2k-1)-(x (2k) + x (2k-2)) / 2 (Formula 2)
Here, x (k) is a pixel value at k points (pixel k on the tile). L means smoothing and H means edge detection.

  When the level 3 two-dimensional discrete wavelet transform is performed according to (Equation 2) described above, the data of the three components is converted into coefficient data of 10 subbands as shown in FIG. FIG. 8 shows the configuration of components converted into coefficient data of 10 subbands (3LL, 3HL, 3LH, 3HH, 2HL, 2LH, 2HH, 1HL, 1LH, 1HH). However, for example, “3HL” indicates a subband obtained by applying a high-pass filter (“H”) to the horizontal component and a low-pass filter (“L”) to the vertical component in stage 3.

  In this way, after each component is converted into a plurality of coefficient data according to the level, if necessary, this is quantized by the quantization unit 30 in units of tiles, and then entropy-encoded by the entropy encoding unit 60. .

  Here, entropy coding is a method in which each coefficient data is divided into bit planes, and bit data existing in each bit plane is encoded by a binary arithmetic encoder (MQ encoder).

  In such a configuration, in order to obtain a reversible code, that is, a code with no image deterioration due to compression, it is necessary to encode all bit planes of all subbands. On the other hand, for example, the compression rate can be increased by discarding and encoding lower bit planes individually for each subband.

  Discarding lower bit planes to increase the compression rate is called truncation, and discarding sub-band bit planes that have little effect on the image quality of the decoded image gives priority to discarding them and optimizes them at a predetermined compression rate. It is possible to obtain a good image quality.

When the subsampled component is wavelet transformed Next, the case where the image data in which the color difference signal is subsampled is encoded will be described below.

  FIG. 3 is a diagram showing pixel ratios in the x direction (horizontal direction) and y direction (vertical direction) of each input / output format. FIG. 3 is a configuration for explaining the pixel ratio in the x direction and y direction of each input / output format, and is not related to the minimum pixel unit of wavelet transform in the prior art.

  In FIG. 3, 4: 2: 2 (horizontal component H = 2, vertical component V = 1: below, expressed as H2V1), 4: 2: 2 (H1V2), and 4: 1: 1 (H1V1). Assume that the components 1 and 2 of each format are subsampled as preprocessing.

  In this configuration, for example, in the case of the 4: 2: 2 (H2V1) format, since sub-sampling is performed in the x direction, data in the x direction is thinned and sub-sampled. Similarly, in the case of 4: 2: 2 (H1V2) format, data in the y direction is thinned out, and in the case of 4: 1: 1 (H1V1), both data in the xy direction are subsampled (thinned out). .

  FIG. 9 shows the state of wavelet coefficients of each component when level 3 two-dimensional discrete wavelet transform is performed on image data in 4: 2: 2 (H2V1) format.

  Referring to FIG. 9, for example, when the tile size is 128 × 128 pixels, the data areas of the subsampled components 1 and 2 are thinned out in the x direction, and thus are 64 × 128 pixels.

When the level 3 two-dimensional discrete wavelet transform is performed on the data in this region, the final subband has half the number of pixels of the horizontal component as compared to the number of pixels of the vertical component, as shown in FIG. The shape is different from the subband of the component which is not subsampled (see FIG. 8).
JP 2000-270335 A JP 2000-134039 A

  As described above, when the two-dimensional discrete wavelet transform is performed on the data in which the X direction and / or the Y direction are thinned, there is a difference in the shape of the subband for each component.

  For this reason, when entropy coding is performed by dividing into bit planes for each subband, '0' or the like is interpolated between pixels for the subsampled bitplane, or the value for the previous pixel is interpolated. Thus, there is a problem that the control according to the existing difference is required and the circuit of the entropy encoding unit 60 becomes complicated.

  Furthermore, when interpolating as described above, the amount of data increases by the amount of interpolation, and there is a problem that the amount of processing increases.

  The present invention has been made in view of the above problems, and an image processing apparatus that can easily generate code data from subsampled image data without increasing a complicated circuit or an amount of data. The purpose is to provide.

  In order to achieve such an object, the present invention has wavelet transform means for performing horizontal and / or vertical wavelet transform on image data in one or more stages, and the wavelet transform means includes the image data. The vertical and / or horizontal wavelet transform is not performed at a predetermined stage.

  As a result, according to the present invention, it is possible to easily generate code data from the subsampled image data without increasing a complicated circuit and an amount of data.

  Further, in the reference invention of the present invention, when the wavelet transform unit thins out a line in a predetermined direction in the image data, the wavelet transform in the predetermined direction is not performed in the predetermined stage, and the line in the image data is performed. If none of these are thinned out, the wavelet transform is performed in all stages.

  Thus, according to the reference invention of the present invention, it is possible to easily generate code data in accordance with the subsampled image data.

  Furthermore, the reference invention of the present invention has encoding means for encoding coefficient data created by the wavelet transform means, and the encoding means does not include a predetermined subband in the coefficient data. The encoding is not performed for the predetermined subband.

  Thereby, in the reference invention of the present invention, it is possible to generate code data having a high compression rate.

  The reference invention of the present invention also includes a one-dimensional horizontal wavelet transform means for performing a horizontal one-dimensional wavelet transform on image data created for each of a plurality of components, and a vertical one for the image data. An image processing apparatus, comprising: a one-dimensional vertical wavelet transform unit for performing a two-dimensional wavelet transform; and a two-dimensional wavelet transform unit comprising the two-dimensional wavelet transform unit, wherein the inverse wavelet transform unit performs two-dimensional wavelet transform in one or more stages. Input / output destination switching means for switching input / output destinations of the one-dimensional horizontal wavelet transform means and / or the one-dimensional vertical wavelet transform means in one stage, and the input / output destination switching means If the line in the specified direction is thinned, the image data To, so as not to perform a one-dimensional wavelet transform with respect to the predetermined direction in the first stage, it is characterized in that switches the output destination.

  As a result, according to the reference invention of the present invention, it is possible to perform appropriate wavelet transform on the subsampled image data, and code data can be easily generated without increasing complicated circuits and data amount. It becomes possible.

  Furthermore, the reference invention of the present invention has input destination switching means for switching the input destination of the subsequent stage of the first stage, and when the input destination switching means has not thinned out any lines of the image data, The input destination is switched so that coefficient data created by performing the two-dimensional wavelet transform in the first stage is input to the subsequent stage.

  As a result, in the reference invention of the present invention, it is possible to distinguish between subsampled image data and non-subsampled image data, and to perform appropriate wavelet transforms corresponding to each of them. Therefore, it is possible to easily generate code data without increasing the number.

  Further, the reference invention of the present invention has inverse wavelet transform means for performing inverse wavelet transform in the horizontal direction and / or vertical direction on coefficient data composed of a plurality of subbands generated by wavelet transform in one or more stages. The image processing apparatus is characterized in that the inverse wavelet transform means does not perform the inverse wavelet transform performed on the coefficient data at a predetermined stage.

  Thus, in the reference invention of the present invention, it is possible to output image data subsampled according to the purpose.

  Furthermore, the reference invention of the present invention is characterized in that the inverse wavelet transform means does not perform inverse wavelet transform on the coefficient data in accordance with the shape of image data to be created.

  As a result, in the reference invention of the present invention, it is possible to easily switch and output subsampled image data and non-subsampled image data.

  The reference invention of the present invention also includes a one-dimensional horizontal inverse wavelet transform means for performing one-dimensional inverse wavelet transform in the horizontal direction on coefficient data comprising one or more subbands, and a vertical direction with respect to the coefficient data. 2D inverse wavelet transform means for performing the one-dimensional inverse wavelet transform and 2D inverse wavelet transform means comprising the two-dimensional inverse wavelet transform means for performing two-dimensional inverse wavelet transform in one or more stages. An image processing apparatus, comprising: an input / output destination switching means for switching input / output destinations of the one-dimensional horizontal reverse wavelet transform means and / or the one-dimensional vertical reverse wavelet transform means in the last stage; The destination switching means applies the coefficient data to the coefficient data according to the shape of the image data to be created. So as not to perform a one-dimensional inverse wavelet transform in the horizontal direction and / or vertically in the last stage, it is characterized in that switches the output destination.

  Thereby, in the reference invention of the present invention, it is possible to output subsampled image data created by performing an appropriate inverse wavelet transform according to the purpose.

  Furthermore, the reference invention of the present invention has an input destination switching means for switching an input destination of the latter stage of the last stage, and the input destination switching means performs two-dimensional inverse wavelet transform by the last stage on the coefficient data. When creating image data that is not applied to the image data, the input destination is switched so that image data that has not been subjected to the two-dimensional inverse wavelet transform in the last stage is input to the subsequent stage.

  Thus, in the reference invention of the present invention, it is distinguished whether the target image data is subsampled image data or non-subsampled image data, and an appropriate inverse wavelet transform is performed accordingly. It becomes possible.

  The reference invention of the present invention is an image processing apparatus having pixel data reading means for reading pixel data for each of a plurality of components from an image, wherein the pixel data reading means sets a component to be input as a pixel unit or The pixel data is read by switching in units of tiles or lines.

  As a result, in the reference invention of the present invention, it is possible to input and output image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, the reference invention of the present invention is characterized in that the pixel data reading means reads pixel data of different components one or more at a time in parallel.

  As a result, in the reference invention of the present invention, not only can the image data be input / output but also the pixel data can be read in parallel so that the buffer required for performing the wavelet transform does not increase. It becomes.

  Further, in the reference invention of the present invention, the plurality of components are first to third components, and the pixel data reading unit thins out vertical lines in the image data of the second and third components. In this case, the component is switched so that the pixel data of the first component, the pixel data of the second component, and the pixel data of the third component are read at a ratio of 2: 1: 1. Yes.

  As a result, in the reference invention of the present invention, it is possible to read image data that has been appropriately subsampled so that a buffer required for performing wavelet transform does not increase.

  Furthermore, the reference invention of the present invention is such that when the pixel data reading means further thins out horizontal lines in the image data of the second and third components, the pixels in the vertical direction of the second and third components It is characterized in that data is read by skipping one pixel.

  As a result, in the reference invention of the present invention, it is possible to read image data that has been appropriately subsampled so that a buffer required for performing wavelet transform does not increase.

  Further, in the reference invention of the present invention, the plurality of components are first to third components, and the pixel data reading means thins out horizontal component lines in the pixel data of the second and third components. In this case, the component is switched so that the pixel data of the first component and the pixel data of the second component or the pixel data of the third component are read at a ratio of 2: 1, and the second component The component and the third component are switched for each line.

  As a result, in the reference invention of the present invention, it is possible to read image data that has been appropriately subsampled so that a buffer required for performing wavelet transform does not increase.

  Furthermore, the reference invention of the present invention is an image processing method including a wavelet transform process for performing horizontal and / or vertical wavelet transform on image data in one or more stages, wherein the wavelet transform process includes the wavelet transform process described above. It is characterized in that vertical and / or horizontal wavelet transform for image data is not performed in a predetermined stage.

  As a result, according to the reference invention of the present invention, it is possible to easily generate code data from the subsampled image data without increasing the number of complicated circuits and the amount of data.

  Furthermore, in the reference invention of the present invention, in the wavelet transform step, when a line in a predetermined direction in the image data is thinned, the wavelet transform in the predetermined direction is not performed in the predetermined stage, and the line in the image data is performed. If none of these are thinned out, the wavelet transform is performed in all stages.

  Thus, according to the reference invention of the present invention, it is possible to easily generate code data in accordance with the subsampled image data.

  Furthermore, the reference invention of the present invention has an encoding step of encoding the coefficient data created in the wavelet transform step, and the encoding step does not include a predetermined subband in the coefficient data. The encoding is not performed for the predetermined subband.

  Thereby, in the reference invention of the present invention, it is possible to generate code data having a high compression rate.

  In addition, the reference invention of the present invention includes an inverse wavelet transform process for performing inverse wavelet transform in the horizontal direction and / or vertical direction in one or more stages on coefficient data composed of a plurality of subbands generated by wavelet transform. An image processing method, wherein the inverse wavelet transform method does not perform an inverse wavelet transform performed on the coefficient data at a predetermined stage.

  Thus, in the reference invention of the present invention, it is possible to output image data subsampled according to the purpose.

  Furthermore, the reference invention of the present invention is characterized in that the inverse wavelet transform step does not perform inverse wavelet transform on the coefficient data according to the shape of the image data to be created.

  As a result, in the reference invention of the present invention, it is possible to easily switch and output subsampled image data and non-subsampled image data.

  The reference invention of the present invention is an image reading method including a pixel data reading step for reading pixel data for each of a plurality of components from an image, wherein the pixel data reading step sets a component to be input as a pixel unit or The pixel data is read by switching in units of tiles or lines.

  As a result, in the reference invention of the present invention, it is possible to input and output image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, the reference invention of the present invention is characterized in that the pixel data reading step reads pixel data of different components one at a time or two or more in parallel.

  As a result, in the reference invention of the present invention, not only can the image data be input / output but also the pixel data can be read in parallel so that the buffer required for performing the wavelet transform does not increase. It becomes.

  Further, in the reference invention of the present invention, the plurality of components are first to third components, and the pixel data reading step thins out vertical lines in the image data of the second and third components. In this case, the component is switched so that the pixel data of the first component, the pixel data of the second component, and the pixel data of the third component are read at a ratio of 2: 1: 1. Yes.

  As a result, in the reference invention of the present invention, it is possible to read image data that has been appropriately subsampled so that a buffer required for performing wavelet transform does not increase.

  Furthermore, in the reference invention of the present invention, when the pixel data reading step further thins out horizontal lines in the image data of the second and third components, the vertical pixels in the second and third components It is characterized in that data is read by skipping one pixel.

  As a result, in the reference invention of the present invention, it is possible to read image data that has been appropriately subsampled so that a buffer required for performing wavelet transform does not increase.

  Further, in the reference invention of the present invention, the plurality of components are first to third components, and the pixel data reading step thins out horizontal component lines in the pixel data of the second and third components. In this case, the component is switched so that the pixel data of the first component and the pixel data of the second component or the pixel data of the third component are read at a ratio of 2: 1, and the second component The component and the third component are switched for each line.

  As a result, in the reference invention of the present invention, it is possible to read image data that has been appropriately subsampled so that a buffer required for performing wavelet transform does not increase.

  As described above, according to the present invention, it is possible to easily generate code data from subsampled image data without increasing a complicated circuit and an amount of data.

[Features of the present invention]
The present invention makes it possible to improve the compression effect over image compression using JPEG-2000 over conventional methods.

  In realizing this, the present invention collects the “0” bits inserted to interpolate the image data thinned out in the prior art to form a so-called “0” bit plane. Here, the “0” bit plane is a bit plane formed with a value of “0”.

  That is, for example, image data composed of three components 0 (= Y), 1 (= Cb), and 2 (= Cr) of Y, Cb, and Cr, and each tile is composed of 128 × 128 pixels. In the case of performing two-dimensional discrete wavelet transform according to JPEG-2000 on image data in which pixels in the y direction in 1 and 2 are thinned to ½ and set to 64 × 128 pixels, the conventional technique uses component 1 , 2, it is necessary to make the same size (128 × 128 pixels) in all the components by interpolating “0” between the pixels.

  On the other hand, in the present invention, a “0” bit plane is added to the components 1 and 2 so that all components have the same size (128 × 128 pixels).

  In JPEG-2000, this “0” bit plane is compressed into 1-bit data. Therefore, in the present invention, since interpolation is performed by adding a “0” bit plane, higher compression is realized.

  However, in the present invention, since the area where the “0” bit plane is formed is determined for each data format, in practice, the process of adding the “0” bit plane is omitted, and the “0” bit plane is set to a predetermined value. The process is executed assuming that the bit plane is formed.

[First Embodiment]
DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, a first embodiment in which the invention is preferably implemented will be described in detail with reference to the drawings.

(Configuration of the first embodiment)
This embodiment is for improving the compression rate by JPEG-2000 in the image processing apparatus shown in FIG.

  In FIG. 1, an encoding / decoding processing unit 2 has a functional block as shown in FIG. 2, and performs color conversion, wavelet conversion, quantization, etc. on input image data. Coding is performed by performing processing such as coefficient order control, coefficient modeling, entropy coding, and the like, and input code data is decoded by performing the reverse processing described above.

  In FIG. 1, the image reading unit 5 includes a CCD camera or the like, and the A / D conversion unit 6 performs A / D conversion on the image data read by the image reading unit 5 and outputs the image data. Is.

  The I / O interface 7 is an interface that mediates transmission / reception of data between the A / D converter 6, the RAM 8, the bus 1, or an external device.

  Further, the CPU 3 controls each unit and the like, and the memory 4 functions as a work area for the CPU 3 and the encoding / decoding processing unit 2.

  In this configuration, the processing operation of the encoding / decoding processing unit 2 will be described in detail below with reference to the drawings.

(Encoding method)
First, an encoding method in the encoding / decoding processing unit 2 will be described with reference to the drawings. However, in the following description, a case where one tile is composed of 128 × 128 pixels will be described as an example. The configuration of the JPEG-2000 encoding functional block in the encoding / decoding processing unit 2 is shown in FIG. However, in the present embodiment, the configuration and operation of the wavelet transform unit 20 are different from those described above, and are as follows.

  In the following description, for convenience, the data format of the image data is 4: 4: 4 format, 4: 2: 2 (H2V1) format, 4: 2: 2 (H1V2) format, 4: 1: 1 (4 : 2: 0) format.

  In the 4: 4: 4 format, the ratio of the number of pixels of each component (for example, Y: Cb: Cr) is 4: 4: 4 (= 1: 1: 1) in RGB or YCbCr format image data. That is, it indicates data that has not been thinned.

  The 4: 2: 2 (H2V1) format indicates data in which the ratio of the number of pixels of each component is 4: 2: 2, and pixels in the vertical direction (y direction) are thinned out. The 2 (H1V2) format indicates data in which the ratio of the number of pixels of each component is 4: 2: 2, and pixels in the horizontal direction (x direction) are thinned out.

  Furthermore, the 4: 1: 1 (H1V1) format indicates data in which the number of pixels of each component is 4: 1: 1 and the pixels in the vertical and horizontal directions are thinned out.

  In the present embodiment, a case will be described in which the image data input to the wavelet transform unit 20 is in the YCbCr format and the wavelet transform to be executed is up to level 3.

When encoding image data subsampled into 4: 2: 2 (H2V1) format First, in this embodiment, a method of encoding image data subsampled into 4: 2: 2 (H2V1) format Will be explained.

  FIG. 5 is a diagram for explaining the wavelet transform executed for the image data in the 4: 2: 2 (H2V1) format for each stage.

  Referring to FIG. 5, since the components 1 and 2 in the image data in the 4: 2: 2 (H2V1) format are subsampled (decimated) in the x direction, the input data is 64 pixels × 128 pixels. Similarly, the output data is 64 pixels × 128 pixels.

  In the present embodiment, when image data of 64 pixels × 128 pixels is input in this way, the wavelet transform unit 20 inputs data that has already been subjected to a low-pass filter in the x direction for the data of components 1 and 2. It is considered that it was done. Therefore, in FIG. 5, only vertical conversion is performed at stage 1. Further, in subsequent stages 2 and 3, the conversion is executed once in both the vertical and horizontal directions.

  By executing the processing in this way, the coefficient data finally obtained for the components 1 and 2 is only 3LL to 2HH and 1LH, and the subband shape is as shown in FIG.

  However, the subband size needs to be the same for all components. Accordingly, in the subbands related to components 1 and 2, the subsequent processing is executed assuming that coefficient data of 1HL and 1HH exists.

  That is, when entropy encoding the coefficient data shown in FIG. 11, the entropy encoding unit 60 considers that the coefficient data of the 1HL and 1HH subbands assumed to exist in advance are all discarded (truncated). Perform encoding. Thereby, in this embodiment, a code compliant with JPEG-2000 is generated.

  In the above method, since the size of each subband region can be considered to be the same in all components, the encoding control method is also the same. Therefore, in the present embodiment, a code conforming to JPEG-2000 can be generated without the need to additionally configure a complicated circuit or the like.

When encoding image data subsampled in 4: 2: 2 (H1V2) format Next, in this embodiment, image data subsampled in 4: 2: 2 (H1V2) format is encoded. A method will be described.

  FIG. 6 is a diagram for explaining the wavelet transform executed for the image data in the 4: 2: 2 (H1V2) format for each stage.

  Referring to FIG. 6, since the components 1 and 2 are sub-sampled (decimated) in the y direction, the input data is 128 pixels × 64 pixels. Similarly, the output data is 128 pixels × 64 pixels.

  In this embodiment, when image data of 128 pixels × 64 pixels is input in this way, the wavelet transform unit 20 inputs data that has already been subjected to a low-pass filter in the y direction for the data of components 1 and 2. It is considered that it was done. Therefore, in FIG. 6, stage 1 performs only horizontal conversion. Further, in subsequent stages 2 and 3, the conversion is executed once in both the vertical and horizontal directions.

  By executing the processing in this manner, the coefficient data finally obtained for the components 1 and 2 is 3LL to 2HH and only 1HL, and the shape of the subband is 1HL in the configuration shown in FIG. Will be replaced with.

  Therefore, in the subsequent processing, in the same manner as described above, it is assumed that there are '0' bit planes of 1LH and 1HH in the subbands related to components 1 and 2, and the coefficient data of 1LH and 1HH are discarded. Entropy encoding is executed to generate a code conforming to JPEG-2000.

  Even in this case, since the size of each subband region can be regarded as the same in all components, the encoding control method is the same, and there is no need to add a complicated circuit or the like. A code compliant with JPEG-2000 can be generated.

When encoding image data subsampled into 4: 1: 1 (H1V1) format In the present embodiment, a method of encoding image data subsampled into 4: 1: 1 (H1V1) format Will be explained.

  FIG. 7 is a diagram for explaining the wavelet transform executed for the image data in the 4: 1: 1 (H1V1) format for each stage.

  Referring to FIG. 7, since the components 1 and 2 are sub-sampled (thinned out) in the x direction and the y direction, the input data (64 pixels × 64 pixels) is halved both vertically and horizontally. Similarly, the output data is output data of 64 pixels × 64 pixels.

  In this embodiment, when image data of 64 pixels × 64 pixels is input in this way, the wavelet transform unit 20 has already applied low-pass filters to the components 1 and 2 in the x and y directions. It is assumed that data has been input, that is, 1 LL data has been input. Therefore, in this embodiment, the conversion is executed once in both the vertical and horizontal directions in the stages 1 and 2, and the third conversion, that is, the stage 3 is omitted.

  By executing the processing in this way, the coefficient data finally obtained for the components 1 and 2 is 3LL to 2HH.

  Therefore, the subsequent processing assumes that there are '0' bit planes of 1HL, 1LH, and 1HH in the subbands related to components 1 and 2, and discards the coefficient data of 1HL, 1LH, and 1HH and entropy Encoding is executed to generate a code compliant with JPEG-2000.

  Even in this case, since the size of each subband region can be regarded as the same in all components, the encoding control method is the same, and there is no need to add a complicated circuit or the like. , A code compliant with JPEG-2000 can be generated.

  As described above, in this embodiment, it is possible to consider that the shape of the subband is the same in any component. Therefore, JPEG is obtained by entropy coding by a certain process without being aware of the format of the input format. It becomes possible to obtain a code compliant with −2000.

  That is, in an image encoding apparatus based on JPEG-2000, for example, when inputting image data such as YCbCr in which color difference data is subsampled, it is only necessary to change the order of wavelet transform for the subsampled Cb and Cr. A code compliant with the JPEG-2000 format can be generated.

(Configuration for executing wavelet transform)
Next, in the present embodiment, a configuration for executing wavelet transform when realizing the above-described encoding method will be described in detail with reference to the block diagram of FIG.

  FIG. 12 is a block diagram illustrating a configuration example of the wavelet transform unit 20 that performs two-dimensional discrete wavelet transform. In FIG. 12, “H” is a block that executes a one-dimensional discrete wavelet transform in the x direction (hereinafter referred to as a horizontal one-dimensional discrete wavelet transform unit), and “V” is a one-dimensional discrete wavelet transform in the y direction. (Hereinafter, referred to as a vertical one-dimensional discrete wavelet transform unit). Furthermore, in FIG. 12, the level 3 two-dimensional discrete wavelet transform is realized by providing three “H” and “V”.

  In addition, “V” and “H” may be interchanged in each stage. Furthermore, in FIG. 12, each selector receives a control signal from a predetermined control unit (for example, CPU 300), and selects which input is output.

When 4: 4: 4 format image data is input In this configuration, when 4: 4: 4 format image data is input, this image data is first converted to a one-dimensional discrete wavelet in the y direction in V101. Conversion is performed. Thereafter, the data output from V101 is input to H103 via the selector 102, and the one-dimensional discrete wavelet transform in the x direction is executed. Thereby, the processing of stage 1 is completed.

  Thereafter, the data output from H103 is input to V105 via the selector 104, and the processing of stage 2 is executed by V105 and H106. Further, the processing of stage 3 is similarly executed by V107 and H108.

  As described above, when 4: 4: 4 format image data is input, in this embodiment, all three sets of “H” and “V” are used for all components in the input image data. The final coefficient data is obtained.

When 4: 2: 2 (H2V1) format image data is input Also, when 4: 2: 2 (H2V1) format image data is input, the above component 4: 4: 4 format is used for component 0 As in the case of, final coefficient data is obtained using all three sets of “H” and “V”.

  For components 1 and 2, y-direction one-dimensional discrete wavelet transform is performed on the input image data in V 101, and this is input to V 105 via the selector 104.

  That is, when image data in the 4: 2: 2 (H2V1) format is input, the selector 104 operates to output the data input from V101.

  Thus, in stage 1 for components 1 and 2, only the one-dimensional discrete wavelet transform in the y direction is executed, and the one-dimensional discrete wavelet transform in the x direction is omitted. Further, the processing in stages 2 and 3 is executed in the same manner as in the 4: 4: 4 format.

  Therefore, when image data in 4: 2: 2 (H2V1) format is input, in this embodiment, all three sets of “H” and “V” are used only for component 0 in the input image data. For components 1 and 2, “H” and “V” other than “H” in stage 1 are used, and final coefficient data is acquired.

When 4: 2: 2 (H1V2) format image data is input Also, when 4: 2: 2 (H1V2) format image data is input, the above-described 4: 4: 4 format is used for component 0 As in the case of, final coefficient data is obtained using all three sets of “H” and “V”.

  For components 1 and 2, the input image data is input to H103 via the selector 102, and the one-dimensional discrete wavelet transform in the x direction is executed.

  That is, when image data in the 4: 2: 2 (H1V2) format is input, this image data is not input to V101 but is input to the selector 102, and the selector 102 outputs this image data to H103.

  Thereby, in stage 1 for components 1 and 2, only the one-dimensional discrete wavelet transform in the x direction is executed, and the one-dimensional discrete wavelet transform in the y direction is omitted. Further, the processing in stages 2 and 3 is executed in the same manner as in the 4: 4: 4 format.

  Therefore, when image data in 4: 2: 2 (H1V2) format is input, in this embodiment, all three sets of “H” and “V” are used only for component 0 in the input image data. For components 1 and 2, “H” and “V” other than “V” in stage 1 are used, and final coefficient data is acquired.

When the image data in the 4: 1: 1 (H1V1) format is input Further, when the image data in the 4: 1: 1 (H1V1) format is input, regarding the component 0, the above 4: 4: 4 format is used. As in the case of, final coefficient data is obtained using all three sets of “H” and “V”.

  In addition, components 1 and 2 are not input to V101 and H103, but are input to H105 via the selector 104, and the processes after stage 2 are executed.

  That is, when image data in the 4: 1: 1 (H1V1) format is input, this image data is input to the selector 104, and the selector 104 outputs this image data to V105.

  Thereby, in stage 1 for components 1 and 2, the process in stage 1 is omitted, and the processes in stages 2 and 3 are executed in the same manner as in the 4: 4: 4 format.

  Therefore, when image data in the 4: 1: 1 (H1V1) format is input, in this embodiment, all three sets of “H” and “V” are used only for the component 0 in the input image data. Then, “H” and “V” after stage 2 are used for the components 1 and 2 to obtain final coefficient data.

  Thus, by adding a simple switching function to the wavelet processing, such as a configuration in which a switch is provided in front of the block of stage 1, subsampled image data and unsubsampled image data are added. Regardless, two-dimensional discrete wavelet transform corresponding to all formats can be realized.

  That is, in an apparatus that performs image coding by JPEG-2000, for example, the order of wavelet transform is changed between image data such as YCbCr in which color difference data is subsampled and image data such as YCbCR and RGB that are not subsampled. This makes it possible to generate a code that conforms to the JPEG-2000 format in a unified manner for both image data.

(Configuration for performing inverse wavelet transform)
Next, in the present embodiment, a configuration for executing the inverse wavelet transform will be described in detail with reference to the block diagram of FIG.

  FIG. 13 is a block diagram illustrating a configuration example of the wavelet transform unit 20 that performs two-dimensional discrete inverse wavelet transform. In FIG. 13, “IH” is a block that executes a one-dimensional discrete inverse wavelet transform in the x direction (hereinafter referred to as a horizontal one-dimensional discrete inverse wavelet transform unit), and “IV” is a one-dimensional discrete in the y direction. This block executes inverse wavelet transform (referred to as a vertical one-dimensional discrete inverse wavelet transform unit). Further, in FIG. 13, three “IH” and “IV” are provided, respectively, so that a level 3 two-dimensional discrete inverse wavelet transform is realized.

  In each stage, “IV” and “IH” may be interchanged. Further, in FIG. 13, each selector receives a control signal from a predetermined control unit (for example, CPU 300) and selects which input is to be output.

  In addition, wavelet coefficient data obtained by a code compliant with the JPEG-2000 standard is input to this configuration.

When outputting 4: 4: 4 format image data In this configuration, when finally outputting 4: 4: 4 format image data, all three sets of “IH” and “IV” are used. Perform a two-dimensional discrete inverse wavelet transform.

  That is, the selector 206 inputs the coefficient data input from the IH 205 to the IV 207. Thereby, on the output side, the image data output from the IV 207 is acquired through all the “IH” and “IV”.

When outputting image data in 4: 2: 2 (H2V1) format On the other hand, when outputting image data in 4: 2: 2 (H2V1) format, the selector 206 outputs the coefficient data output from the IV 204. Is output to IV207. Thereby, on the output side, the IH 205, that is, the one-dimensional discrete inverse wavelet transform in the x direction in the stage 3 is omitted, and the image data is acquired.

When outputting image data in 4: 2: 2 (H1V2) format When outputting image data in 4: 2: 2 (H1V2) format, the selector 206 outputs the data output from the IH205, On the output side, the data output from the selector 206 is acquired. Thereby, on the output side, IV207, that is, the one-dimensional discrete inverse wavelet transform in the y direction in stage 3, is omitted, and image data is acquired.

When outputting image data in the 4: 1: 1 (H1V1) format Further, when outputting image data in the 4: 1: 1 (H1V1) format, the data output from the IV 204 is acquired on the output side. Thereby, on the output side, the IH 205 and IV 207, that is, the two-dimensional discrete inverse wavelet transform in the x direction and the y direction in the stage 3 is omitted, and image data is acquired.

  In this way, by providing a switch in front of the block of stage 3 and switching the conversion of stage 3 according to the format of the image data to be acquired, two-dimensional discrete inverse wavelet transform corresponding to all formats is realized, It is possible to switch between output of image data such as YCbCr in which color difference data is subsampled and output of image data that is not subsampled.

(Image data input / output method)
An image data input / output method suitable for inputting image data to the encoding function block configured as shown in FIG. 2 will be described below in detail with reference to the drawings. This image data input / output method operates in the image reading unit 5.

  Generally, in the NTSC system or the like, the YCbCr signal is input / output in order while switching the components for each pixel. However, in the digital camera or the like, the order of the components is switched in units of frames and the YCbCr signal is input / output. It has become.

  However, when the encoding function block shown in FIG. 2 is configured to input and output YCbCr signals by switching the order of components in units of frames, the size required for a buffer for temporarily storing image data is very large. Thus, the chip size of the LSI on which this is mounted becomes enormous.

  Therefore, in the present embodiment, in order to efficiently perform wavelet transform on the input image data, the image data is input by the following method (image data input / output method) and output to the encoding function block. .

  In the image data input / output method according to the present embodiment, as shown in FIG. 14, first, the image data is divided into reference image units (tiles). Thereafter, the image data divided into the tiles is sequentially input for each tile, and is output to the encoding function block shown in FIG.

  For each tile, as shown in FIG. 15, for example, image data is input in raster order with the upper left as a start point and the lower right as an end point, and this is output.

  Here, (Table 1) illustrates a list of image data formats input and output in the present embodiment.

（表１）に示すように、本実施形態では、各フォーマット（4:4:4，4:2:2(H2V1)，4:2:2(H1V2)，4:1:1(4:2:0)）に対して、例えば８ビット又は１０ビットの色形式が対応付けられている。また、色形式としては、ＲＧＢ形式又はＹＣｂＣｒ形式が用いられている。

As shown in Table 1, in this embodiment, each format (4: 4: 4, 4: 2: 2 (H2V1), 4: 2: 2 (H1V2), 4: 1: 1 (4: 2 : 0)) is associated with, for example, an 8-bit or 10-bit color format. As the color format, the RGB format or the YCbCr format is used.

・ 4: 4: 4 format (RGB, YCbCr)
Among these, the operation when inputting / outputting image data in the 4: 4: 4 format RGB format or YCbCr format in time series will be described with reference to FIG.

  In (a) of FIG. 16, when inputting / outputting image data in the 4: 4: 4 format RGB format or YCbCr format, the pixels in the tile are converted into components (R, G, B, or Y, Cb, Cr). ) Are input while switching sequentially, and the operation is performed to output this.

  That is, in the present embodiment, the processing for each component (R, G, B, or Y, Cb, Cr) is executed without being separated.

-4: 2: 2 (H2V1) format (YCbCr)
The operation when inputting / outputting image data in the 4: 2: 2 (H2V1) format in the YCbCr format in time series will be described with reference to FIG.

  In (b) of FIG. 16, when inputting / outputting image data in the 4: 2: 2 (H2V1) format YCbCr format, the processing for each component of Y and Cb or Cr is separated and executed. .

  That is, in FIG. 16B, the tile for the process for component 1 (= Y) is a 128 × 128 pixel tile, but in the process for component 1 (= Cb), 2 (= Cr), The tile is a 64 × 128 pixel tile.

-4: 2: 2 (H1V2) format (YCbCr)
Similarly, when inputting / outputting image data in the 4: 2: 2 (H1V2) format YCbCr format, in the diagram shown in (c) of FIG. 16, the tile is 128 in the process for the component 0 (= Y). The tiles are set to 128 × 64 pixels in the processing for the components 1 (= Cb) and 2 (= Cr).

  Accordingly, even in this case, the processes for the components Y and Cb or Cr are separated and executed.

・ 4: 1: 1 (4: 2: 0) format (YCbCr)
Furthermore, when inputting / outputting image data in the YCbCr format of the 4: 1: 1 (4: 2: 0) format, in the diagram shown in FIG. 16 (d), the tile is processed by the process for the component 0 (= Y). The tiles are set to 128 × 128 pixels, and the tiles are set to 64 × 64 pixels by the processing for the components 1 (= Cb) and 2 (= Cr).

  Accordingly, even in this case, the processes for the components Y and Cb or Cr are separated and executed.

  As described above, in this embodiment, when inputting / outputting image data that has not been thinned out, the processing for each component is executed without being separated, and when inputting / outputting image data in which pixels for Cb and Cr are thinned out. The process for the Y component and the process for the Cb or Cr component are executed separately. Thereby, an image data input / output method appropriate for each format is realized.

(Specific example of image data input / output method)
Next, a specific example of the image data input / output method for each input / output image format shown in (Table 1) will be given using (Table 2) to (Table 13) shown below, and this will be described in time series. .

  However, in each table shown below, pixel data is composed of R (red), G (green), B (blue), Y (illuminance), and Cb · Cr (color difference).

  In the table, “clock #” indicates the number of clocks from the start of pixel data input (output), and “cycle #” indicates the cycle of repeated processing according to the clock. “Pixel #” indicates the number of the pixel to be processed (the number assigned to each pixel in the tile shown in FIG. 16; however, regarding component 0 (= Y)), and pixel data (or pixel data 3). , 2, 1, 0) indicate values when each pixel is read for each component.

  Further, in the table, “x” in the pixel data indicates that there is no input / output of pixel data, and “n” in the description is an arbitrary positive integer.

  ・ 4: 4: 4 format (RGB, YCbCr) input / output (serial)

（表２）は、４：４：４フォーマット（ＲＧＢ，ＹＣｂＣｒ）の画素データを画素順にシリアルに入出力する際の流れを時系列に従って説明するための表である。

(Table 2) is a table for explaining the flow when serially inputting / outputting pixel data of 4: 4: 4 format (RGB, YCbCr) in pixel order in chronological order.

  Referring to Table 2, the illuminance “Yn” (or “Rn”) of the pixel “n” is input (output) at the cycle “0”, and the color difference “Cbn” (at the cycle “1”) of the pixel “n”. Or “Gn”) is input, and the color difference “Crn” (or “Bn”) of the pixel “n” is input (output) in cycle “2”. In Table 2, this series of operations is one cycle. This cycle is repeated until completion for all pixels (for example, 16383 pixels for a 128 × 128 pixel tile).

  As described above, in (Table 2), pixel data of one pixel is input / output in three clocks.

  -4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) input / output (serial)

また、（表３）は、４：２：２（Ｈ２Ｖ１）又は４：１：１フォーマット（ＹＣｂＣｒ）の画素データを画素順にシリアルに入出力する際の流れを時系列に従って説明するための表である。但し、（表３）は、画素単位でＣｂ，Ｃｒが交互に入出力されるよう構成した場合の流れを示している。

Further, (Table 3) is a table for explaining the flow when serially inputting / outputting pixel data of 4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) in pixel order in time series. is there. However, (Table 3) shows a flow in a case where Cb and Cr are alternately input / output in units of pixels.

  Referring to (a) in (Table 3), the color difference “Cbn” of the pixel “n” is input (output) in the cycle “0”, and the illuminance “Yn” of the pixel “n” is input (output) in the cycle “1” ( The color difference “Crn” of the pixel “n” is input (output) in the cycle “2”, and the illuminance “Yn + 1” of the pixel “n + 1” is input (output) in the cycle “3”. In Table 3, this series of operations is one cycle. In addition, this cycle includes all pixels (for example, if component 0 (= Y) is a tile of 128 × 128 pixels, 16383 pixels (however, 8191 pixels of 64 × 128 pixels for components 1 and 2)) ) Until it is completed.

  As described above, in (Table 3), pixel data of two pixels is input / output in four clocks.

  However, when pixel data in the 4: 1: 1 format is read, as shown in (b) of (Table 3), pixels to be performed on the even-numbered line from the top in the line in the component 0 (= Y) The pixel data of the color differences Cb and Cr are not input (output) during data input / output.

  -4: 2: 2 (H1V2) format (YCbCr) input / output (serial)

また、（表４）は、４：２：２（Ｈ１Ｖ２）フォーマット（ＹＣｂＣｒ）の画素データを画素順にシリアルに入出力する際の流れを時系列に従って説明するための表である。但し、（表４）は、ライン単位でＣｂ，Ｃｒが交互に入出力されるよう構成した場合の流れを示している。

Further, (Table 4) is a table for explaining the flow when serially inputting / outputting pixel data in the 4: 2: 2 (H1V2) format (YCbCr) in the order of pixels in time series. However, (Table 4) shows a flow in a case where Cb and Cr are alternately input / output in line units.

  In the odd-numbered line in the component 0 (= Y), as shown in (a) of (Table 4), for example, the color difference “Cbn” of the pixel “n” is input (output) in cycle “0”, and the cycle “1” inputs (outputs) the illuminance “Yn” of the pixel “n”. In addition, in the odd-numbered line in the component 0 (= Y), as shown in (b) of (Table 4), for example, cycle “0” (for example, clock “256”: where 1 tile is 128 × 128 pixels) The color difference “Crn (Cr0)” of the pixel “n (0)” is input (output), and the illuminance “Yn + 127 (Y128) of the pixel“ n + 127 (128) ”in the cycle“ 1 ”(clock“ 257 ”). ) 'Is input (output).

  Thus, in (Table 4), two clocks are set as one cycle, and input / output color differences Cb and Cr are exchanged for each line.

  In addition, this cycle includes all pixels (for example, if component 0 (= Y) is a tile of 128 × 128 pixels, 16383 pixels (however, for components 1 and 2 are 8191 pixels of 128 × 64 pixels)) ) Until it is done.

  As described above, in (Table 4), pixel data of one pixel is input / output in two clocks.

  -4: 2: 2 (H2V1, H1V2) format (YCbCr) input / output (serial)

また、（表５）は、４：２：２（Ｈ２Ｖ１，Ｈ１Ｖ２）フォーマット（ＹＣｂＣｒ）の画素データをタイル順にシリアルに入出力する際の流れを時系列に従って説明するための表である。但し、（表５）は、タイル単位でＹを入出力し、次にタイル単位でＣｂ，Ｃｒを交互に入出力するよう構成した場合の流れを示している。

(Table 5) is a table for explaining the flow when inputting / outputting pixel data in 4: 2: 2 (H2V1, H1V2) format (YCbCr) serially in the tile order in time series. However, (Table 5) shows a flow in a case where Y is input / output in tile units and then Cb and Cr are input / output alternately in tile units.

  That is, in (a) of (Table 5), first, the illuminance signal Y ('Y0' to 'Y16383') for one tile is input / output by the operation from the clock '0' to '16383', and then In (b) of (Table 5), the color difference signals Cb and Cr ('Cb0' to 'Cb8191', 'Cr0' to 'Cr8191') for one tile are obtained by the operations from clocks '16384' to '32767'. Are alternately input and output.

  Accordingly, in (Table 5), pixel data of one pixel is input / output in two clocks.

  ・ 4: 4: 4 format (RGB, YCbCr) input / output (2 data parallel)

（表６）は、４：４：４フォーマット（ＲＧＢ，ＹＣｂＣｒ）の画素データを画素順に２つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。

(Table 6) is a table for explaining the flow when inputting / outputting pixel data of 4: 4: 4 format (RGB, YCbCr) in parallel in pixel order for each two pixel data in time series.

  Referring to (Table 6), the illuminance “Yn” (or “Rn”) and the color difference “Cbn” (or “Gn”) of the pixel “n” are input (output) in the cycle “0”, and the cycle “1”. The color difference “Crn” (or “Bn”) of the pixel “n” is input. In Table 6, this series of operations is one cycle. This cycle is repeated until completion for all pixels (for example, 16383 pixels for a 128 × 128 pixel tile).

  As described above, in (Table 6), pixel data of one pixel is input / output in two clocks.

  -4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) input / output (2 data parallel)

また、（表７）は、４：２：２（Ｈ２Ｖ１）又は４：１：１フォーマット（ＹＣｂＣｒ）の画素データを画素順に２つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。但し、（表７）は、画素単位でＣｂ，Ｃｒが交互に入出力されるよう構成した場合の流れを示している。

(Table 7) explains the flow when inputting / outputting pixel data in 4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) in parallel in order of two pieces of pixel data in order of pixels. It is a table for doing. However, (Table 7) shows a flow in a case where Cb and Cr are alternately input / output in units of pixels.

  Referring to (a) of (Table 7), the illuminance “Yn” and the color difference “Cbn” of the pixel “n” are input (output) in the cycle “0”, and the illuminance of the pixel “n + 1” in the cycle “1”. The color difference 'Crn' between Yn + 1 'and pixel' n 'is input (output). In Table 7, this series of operations is one cycle. In addition, this cycle includes all pixels (for example, if component 0 (= Y) is a tile of 128 × 128 pixels, 16383 pixels (however, 8191 pixels of 64 × 128 pixels for components 1 and 2)) ) Until it is completed.

  As described above, in (Table 7), input / output of pixel data of two pixels is performed in four clocks.

  However, when pixel data in the 4: 1: 1 format is read out, as shown in (b) of (Table 7), in the line in the component 0 (= Y), the pixels to be applied to the even-numbered line from the top The pixel data of the color differences Cb and Cr are not input (output) during data input / output.

  ・ Input and output of 4: 2: 2 (H1V2) format (YCbCr) (2 data parallel)

また、（表８）は、４：２：２（Ｈ１Ｖ２）フォーマット（ＹＣｂＣｒ）の画素データを画素順に２つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。但し、（表８）は、ライン単位でＣｂ，Ｃｒが交互に入出力されるよう構成した場合の流れを示している。

Further, (Table 8) is a table for explaining the flow when inputting / outputting pixel data in the 4: 2: 2 (H1V2) format (YCbCr) in parallel in order of each pixel data in the order of pixels. . However, (Table 8) shows a flow in a case where Cb and Cr are alternately input / output in line units.

  In the odd-numbered line in the component 0 (= Y), as shown in (a) of (Table 8), for example, the illuminance “Yn” and the color difference “Cbn” of the pixel “n” are input (cycle “0”) ( Output). In addition, in the odd-numbered line in the component 0 (= Y), as shown in (b) of (Table 8), for example, cycle “0” (for example, clock “256, 257”: where 1 tile is 128 × The illuminance “Yn + 127 (Y128)” of the pixel “n + 127 (128)” and the color difference “Crn (Cr0)” of the pixel “n (0)” are input (output).

  Thus, in (Table 8), two clocks are set as one cycle, and the input / output color differences Cb and Cr are exchanged for each line.

  In addition, this cycle includes all pixels (for example, if component 0 (= Y) is a tile of 128 × 128 pixels, 16383 pixels (however, for components 1 and 2 are 8191 pixels of 128 × 64 pixels)) ) Until it is done.

  As described above, in (Table 8), pixel data of one pixel is input / output in two clocks.

  ・ 4: 4: 4 format (RGB, YCbCr) input / output (3 data parallel)

（表９）は、４：４：４フォーマット（ＲＧＢ，ＹＣｂＣｒ）の画素データを画素順に３つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。

(Table 9) is a table for explaining the flow when inputting / outputting pixel data of 4: 4: 4 format (RGB, YCbCr) in parallel in order of pixels for every three pieces of pixel data in time series.

  Referring to (Table 9), the illuminance “Yn” (or “Rn”) and the color difference “Cbn” (or “Gn”) and the color difference “Crn” (or “Bn”) of the pixel “n” in the cycle “0”. Is entered. In (Table 9), this series of operations is one cycle. This cycle is repeated until completion for all pixels (for example, 16383 pixels for a 128 × 128 pixel tile).

  As described above, in (Table 9), pixel data of one pixel is input / output in two clocks.

  -4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) input / output (4 data parallel)

また、（表１０）は、４：２：２（Ｈ２Ｖ１）又は４：１：１フォーマット（ＹＣｂＣｒ）の画素データを画素順に４つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。但し、（表１０）は、画素単位でＣｂ，Ｃｒが交互に入出力されるよう構成した場合の流れを示している。

(Table 10) explains the flow when inputting / outputting pixel data in 4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) in parallel in pixel order for each of the four pixel data in time series. It is a table for doing. However, (Table 10) shows a flow in a case where Cb and Cr are alternately input / output in units of pixels.

  Referring to (a) of (Table 10), the illuminance 'Yn' and the color difference 'Cbn' and the color difference 'Crn' of the pixel 'n' and the illuminance 'Yn + 1' of the pixel 'n + 1' are input in the cycle '0' ( Output). In Table 10, this series of operations is one cycle. In addition, this cycle includes all pixels (for example, if component 0 (= Y) is a tile of 128 × 128 pixels, 16383 pixels (however, 8191 pixels of 64 × 128 pixels for components 1 and 2)) ) Until it is completed.

  As described above, in (Table 10), pixel data of two pixels is input / output in four clocks.

  However, when pixel data in the 4: 1: 1 format is read out, as shown in (b) of (Table 10), in the line in the component 0 (= Y), the pixels to be applied to the even-numbered line from the top The pixel data of the color differences Cb and Cr are not input (output) during data input / output.

  -4: 2: 2 (H1V2) format (YCbCr) input / output (4 data parallel)

また、（表１１）は、４：２：２（Ｈ１Ｖ２）フォーマット（ＹＣｂＣｒ）の画素データを画素順に４つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。但し、（表１１）は、ライン単位でＣｂ，Ｃｒが交互に入出力されるよう構成した場合の流れを示している。

Further, (Table 11) is a table for explaining the flow when inputting / outputting pixel data of 4: 2: 2 (H1V2) format (YCbCr) in parallel in order of pixels for every four pixel data in time series. . However, (Table 11) shows a flow in a case where Cb and Cr are alternately input / output in line units.

  In the odd-numbered line in component 0 (= Y), as shown in (a) of (Table 11), for example, in cycle “0”, the illuminance “Yn” of pixel “n” and the color difference “Cbn” and pixel “ An illuminance “Yn + 1” of n + 1 ”and a color difference“ Cbn + 1 ”are input (output). In addition, in the odd-numbered line in the component 0 (= Y), as shown in (b) of (Table 11), for example, cycle “0” (for example, clocks “256, 257, 258, 259”: The illuminance 'Yn + 127 (Y128)' of the pixel 'n + 127 (128)' and the color difference 'Crn (Cr0)' and the pixel 'n + 128 (129)' of the pixel 'n (127)' and the pixel 'n (0)' The illuminance “Yn + 128 (Y129)” and the color difference “Crn + 1 (Cr1)” of the pixel “n + 1 (1)” are input (output).

  Thus, in (Table 11), four clocks are set as one cycle, and input / output color differences Cb and Cr are exchanged for each line.

  In addition, this cycle includes all pixels (for example, if component 0 (= Y) is a tile of 128 × 128 pixels, 16383 pixels (however, for components 1 and 2 are 8191 pixels of 128 × 64 pixels)) ) Until it is done.

  Thus, in (Table 11), input / output of pixel data of two pixels is performed in four clocks.

  -Grayscale format input / output (serial)

（表１２）は、グレイスケールフォーマットの画素データを画素順にシリアルに入出力する際の流れを時系列に従って説明するための表である。

(Table 12) is a table for explaining the flow when serially inputting / outputting pixel data in the gray scale format in the pixel order in time series.

  Referring to (Table 12), the illuminance “Yn” of the pixel “n” is input (output) in cycle “0”. In Table 12, this operation is one cycle. This cycle is repeated until completion for all pixels (for example, 16383 pixels for a 128 × 128 pixel tile).

  As described above, in (Table 12), input / output of pixel data of one pixel is performed in two clocks.

  -Grayscale format input / output (2 data parallel)

（表１３）は、グレイスケールフォーマットの画素データを画素順に２つの画素データ毎パラレルに入出力する際の流れを時系列に従って説明するための表である。

(Table 13) is a table for explaining the flow when inputting / outputting pixel data in gray scale format in parallel in the order of pixels for every two pieces of pixel data in time series.

  Referring to (Table 13), the illuminance “Yn” of the pixel “n” and the illuminance “Yn + 1” of the pixel “n + 1” are input in the cycle “0”. In Table 13, this operation is one cycle. This cycle is repeated until completion for all pixels (for example, 16383 pixels for a 128 × 128 pixel tile).

  As described above, in (Table 13), pixel data of two pixels is input / output in four clocks.

  By performing input / output in the above-described procedure, input / output processing is sequentially performed on a pixel-by-pixel basis for image data in 4: 4: 4 format (RGB, YCbCr), and image data in other formats. On the other hand, pixel data can be input / output in units of tiles by separating illuminance (Y) and color difference (Cb, Cr).

  Further, as shown in the above (Table 6) to (Table 11) and (Table 13), in the present embodiment, a plurality of pixel data can be processed in parallel.

(Timing chart example)
Next, a timing chart when inputting / outputting pixel data of 4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) shown in the above (Table 7) in parallel every two pixel data in pixel order. An example will be described with reference to FIG.

  Referring to FIG. 17, in the operation for inputting / outputting pixel data of 4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) in parallel in order of each pixel data, four clocks are one cycle. One cycle is divided into two cycles of cycle “0” and cycle “1”.

  Therefore, first, when the cycle “0” is started, the illuminance “Y0” and the color difference “Cb0” of the pixel “0” are input (output) as the pixel data 1, 0. Next, when the cycle “1” is started, the illuminance “Y1” of the pixel “1” and the color difference “Cr0” of the pixel “0” are input (output) as the pixel data 1 and 0. This is repeated in subsequent operations.

  Further, for other tables, a timing chart as shown in FIG. 17 can be created based on each table.

  By inputting image data in the image reading unit 5 in such a procedure and inputting it to the encoding / decoding processing unit 2 (encoding function block), an efficient encoding process is realized in this embodiment. It becomes possible to do.

[Other Embodiments]
Further, each of the above-described embodiments is merely an example of a preferred embodiment of the present invention, and the present invention can be implemented with various modifications without departing from the gist thereof.

  As described above, according to the present invention, it is possible to easily generate code data from subsampled image data without increasing a complicated circuit and an amount of data.

  Furthermore, according to the reference invention of the present invention, code data can be easily generated according to subsampled image data.

  Furthermore, according to the reference invention of the present invention, it is possible to generate code data having a high compression rate.

  Further, according to the reference invention of the present invention, it is possible to perform an appropriate wavelet transform on the subsampled image data, and it is possible to easily generate code data without increasing complicated circuits and data amount. It becomes possible to do.

  Furthermore, according to the reference invention of the present invention, it is possible to distinguish between subsampled image data and non-subsampled image data, and to perform appropriate wavelet transforms corresponding to each of them. It is possible to easily generate code data without increasing the amount.

  Further, according to the reference invention of the present invention, it is possible to output image data subsampled according to the purpose.

  Furthermore, according to the reference invention of the present invention, it is possible to easily switch and output subsampled image data and non-subsampled image data.

  Further, according to the reference invention of the present invention, it is possible to output subsampled image data generated by performing an appropriate inverse wavelet transform according to the purpose.

  Furthermore, according to the reference invention of the present invention, it is distinguished whether the target image data is subsampled image data or non-subsampled image data, and an appropriate inverse wavelet transform is performed accordingly. Can be applied.

  In addition, according to the reference invention of the present invention, it is possible to input and output image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible not only to input / output image data but also to read pixel data in parallel so that a buffer required for performing wavelet transform does not increase. It becomes possible.

  Furthermore, according to the reference invention of the present invention, it is possible to read appropriately subsampled image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible to read appropriately subsampled image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible to read appropriately subsampled image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible to easily generate code data from subsampled image data without increasing a complicated circuit or an amount of data.

  Furthermore, according to the reference invention of the present invention, code data can be easily generated according to subsampled image data.

  Furthermore, according to the reference invention of the present invention, it is possible to generate code data having a high compression rate.

  Further, according to the reference invention of the present invention, it is possible to output image data subsampled according to the purpose.

  Furthermore, according to the reference invention of the present invention, it is possible to easily switch and output subsampled image data and non-subsampled image data.

  In addition, according to the reference invention of the present invention, it is possible to input and output image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible not only to input / output image data but also to read pixel data in parallel so that a buffer required for performing wavelet transform does not increase. It becomes possible.

  Furthermore, according to the reference invention of the present invention, it is possible to read appropriately subsampled image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible to read appropriately subsampled image data so that a buffer required for performing wavelet transform does not become large.

  Furthermore, according to the reference invention of the present invention, it is possible to read appropriately subsampled image data so that a buffer required for performing wavelet transform does not become large.

1 is a block diagram illustrating a configuration example of an image processing device according to a first embodiment of the present invention. It is a block diagram which shows the structural example of the encoding functional block of the encoding / decoding process part 2 by the 1st Embodiment of this invention. Examples of pixel ratios in the x and y directions in 4: 4: 4, 4: 2: 2 (H2V1), 4: 2: 2 (H1V2), 4: 1: 1, grayscale input / output formats FIG. It is a figure for demonstrating the two-dimensional discrete wavelet transformation performed with respect to the image data of 4: 4: 4 format for every stage. It is a figure for demonstrating for each stage two-dimensional discrete wavelet transformation performed with respect to the image data of 4: 2: 2 (H2V1) format in the 1st Embodiment of this invention. It is a figure for demonstrating the two-dimensional discrete wavelet transformation performed with respect to the image data of 4: 2: 2 (H1V2) format for every stage in the 1st Embodiment of this invention. It is a figure for demonstrating the two-dimensional discrete wavelet transformation performed with respect to the image data of a 4: 1: 1 format in the 1st Embodiment of this invention for every stage. It is a figure for demonstrating the example of the shape of the subband produced | generated by performing the level 3 two-dimensional discrete wavelet transform with respect to the component which is not subsampled. It is a figure for demonstrating the two-dimensional discrete wavelet transformation performed with respect to the image data of 4: 2: 2 (H2V1) format for every stage in a prior art. It is a figure for demonstrating the example of a shape of the subband produced | generated by the two-dimensional discrete wavelet transform shown in FIG. It is a figure for demonstrating the system regulations of the subband produced | generated by the two-dimensional discrete wavelet transform shown in FIG. It is a block diagram which shows the structural example of the wavelet transformation part 20 which performs the wavelet transformation by the 1st Embodiment of this invention. It is a block diagram which shows the structural example of the wavelet transformation part 20 which performs the inverse wavelet transformation by the 1st Embodiment of this invention. It is a block diagram which shows the structure of the tile produced | generated by dividing image data in the image data input / output method by the 1st Embodiment of this invention. It is a figure for demonstrating the order which reads the pixel data in each tile divided | segmented in FIG. It is a figure for demonstrating the operation | movement at the time of inputting / outputting image data according to the time series in the 1st Embodiment of this invention, (a) is image data by the RGB format or YCbCr format of 4: 4: 4 format. It is a figure for demonstrating the operation | movement at the time of inputting / outputting, (b) is a figure for demonstrating the operation | movement at the time of inputting / outputting the image data by YCbCr format of 4: 2: 2 (H2V1) format, (C) is a figure for demonstrating the operation | movement at the time of inputting / outputting the image data by the 4: 2: 2 (H1V2) format YCbCr format, (d) is 4: 1: 1 (4: 2: 0). It is a figure for demonstrating the operation | movement at the time of inputting / outputting the image data of a format YCbCr format. 8 is a timing chart showing a flow when pixel data in 4: 2: 2 (H2V1) or 4: 1: 1 format (YCbCr) shown in (Table 7) is input / output in parallel for each pixel data in pixel order. .

Explanation of symbols

1 Bus 2 Encoding / Decoding Processing Unit 3 CPU
4 Memory 5 Image Reading Unit 6 A / D Conversion Unit 7 I / O Interface 8 RAM
DESCRIPTION OF SYMBOLS 10 Color conversion part 20 Wavelet conversion part 30 Quantization part 40 Coefficient order control part 50 Coefficient modeling part 60 Entropy encoding part 101,105,107 Vertical direction one-dimensional discrete wavelet conversion part (V)
102, 104, 206 Selectors 103, 106, 108 Horizontal one-dimensional discrete wavelet transform unit (H)
201, 203, 205 Horizontal one-dimensional discrete inverse wavelet transform unit (IH)
202,204,207 Vertical one-dimensional discrete inverse wavelet transform unit (IV)

Claims (1)

Wavelet transform in the horizontal direction and / or vertical direction is performed in one or more stages for each component of the image data composed of a plurality of components, and subband coefficient data output for each stage is converted into a bit plane for each component. An image encoding device that divides and encodes
Wavelet transform means for performing the wavelet transform;
Encoding means for encoding the coefficient data created in the wavelet transform means,
The wavelet transform means does not perform wavelet transform in the direction of the subsample of the component that is subsampled among the components at a predetermined stage,
The encoding means discards the coefficient data in the direction of the subsample in a stage where the wavelet transform is not performed as a zero bit plane, generates and encodes data representing the zero bit plane,
Furthermore, in the stage, when inputting and outputting image data in a predetermined format,
If the input / output image data is not thinned out, the process for the component is executed without separation. If the input / output image data is thinned out, the process for the component is executed separately. An image encoding apparatus characterized by that .

Images