JPWO2018230465A1 - Noise removal processing system, noise removal processing circuit, and noise removal processing method - Google Patents

Abstract

The storage unit 81 stores a plurality of lines of pixels received in units of lines. The first wavelet transform unit 82 reads pixels of a plurality of lines stored in the storage unit 81, performs wavelet transform on each line included in the plurality of lines at the same timing, and includes a plurality of first high frequency components and A plurality of first low frequency components are generated. The first inverse wavelet transform means 86 performs inverse wavelet transform based on the second high frequency component and the second low frequency component from which the noise component is degenerated, and generates a first inverse transform signal. The second inverse wavelet transform unit 87 performs inverse wavelet transform on the basis of the plurality of first high frequency components with the noise components degenerated and the first inverse transform signal, and generates a second inverse transform signal. .

Description

  The present invention relates to a noise removal processing system, a noise removal processing method, and a noise removal processing circuit, and more particularly, a noise removal processing system, a noise removal processing circuit, and a noise removal processing method for removing noise in an image by using a plurality of wavelet transforms. About.

  In a data compression process or a noise removal process for a digital signal such as an image or sound, a process for converting a discrete signal into a frequency domain is often used. For example, JPEG (Joint Photographic Experts Group), which is one of the methods for compressing images, performs discrete cosine transform, which is a process of transforming the spatial signal of the image into the frequency domain, and then the amount of information is obtained by quantization. The amount of data is reduced by performing entropy coding after the reduction.

  In the compression processing by JPEG, compression of data having less influence on the original image is realized by performing quantization after concentrating main energy included in a signal in a low frequency region by discrete cosine transform.

  In JPEG 2000, which is a standard following JPEG, a compression method using discrete wavelet transform is employed. Hereinafter, the discrete wavelet transform is simply referred to as wavelet transform since it is intended for digital signals. In the wavelet transform, pixel values for several pixels are input, and the pixel value of the input image is separated into a low frequency component and a high frequency component, respectively.

  In the two-dimensional wavelet transform, a horizontal wavelet transform and a vertical wavelet transform are sequentially performed. When the low frequency component is separated into the low frequency component and the high frequency component in the horizontal direction and the vertical direction, the low frequency component relatively retains the color information of the original image. On the other hand, in a region including a high frequency, a portion where the pixel value changes abruptly in an image, that is, a vibration component derived from object edge information or noise information is held. By applying this also in the vertical direction, the two-dimensional image is separated into a low frequency component and a high frequency component in the horizontal direction and the vertical direction, respectively.

  FIG. 16 is an explanatory diagram illustrating a state in which the two-dimensional wavelet transform is applied to a two-dimensional image. When wavelet transform is applied to 2 × 2 4 pixels, it is separated into a low frequency component of 1 pixel and a high frequency component of 3 pixels. Here, the two-dimensional image is separated into the first high-frequency component, the second high-frequency component, and the low-frequency component by accumulating only the low-frequency components and performing wavelet transform only on the low-frequency components again. . In this way, performing wavelet transform a plurality of times and converting it to a plurality of frequency bands is called multi-resolution analysis by wavelet transform.

  Non-Patent Document 1 describes an image noise removal processing method using wavelet transform. FIG. 17 is an explanatory diagram illustrating an image noise removal processing method. In the image noise removal processing method shown in FIG. 17, processing is performed by separating into a plurality of frequency bands by multi-resolution analysis by wavelet transform and degenerating noise included in high-frequency components in each band. This process is called wavelet shrinkage. Thereafter, inverse wavelet transform is performed on the high-frequency component and low-frequency component in which noise is degenerated, thereby restoring the original image information.

  When data compression or noise removal processing is performed on an image signal, particularly a moving image signal, these processing are executed on an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array) mounted with a dedicated circuit. May be. This is because the photographing apparatus realizes high-speed processing and further real-time processing even when a high-performance CPU (Central Processing Unit) cannot be mounted due to power, mounting area, or the like. In the hardware as described above, since the mounting area is limited, stream processing for receiving input in units of lines may be performed without arranging a memory for holding an image frame inside.

  Here, Patent Document 1 describes a configuration example of a system that performs two-dimensional wavelet transform by stream processing. In the system described in Patent Document 1, input is received in the order of raster scan, and the first wavelet transform is performed. The low frequency component output from the first wavelet transform is accumulated in the memory (line buffer), and the second frequency component is stored in the line buffer when the low frequency component necessary for executing the second wavelet transform is accumulated in the line buffer. Wavelet transform is performed.

  On the other hand, Patent Document 2 describes a noise removal processing system that can remove noise contained in a signal while suppressing an increase in the amount of memory to be used. Specifically, in Patent Document 2, in addition to the above-described line buffer for wavelet transform, high-frequency components are accumulated until the first inverse wavelet transform is performed after the first wavelet transform is performed. It is stated that the memory is required.

International Publication No. 2014/077245 International Publication No. 2013/145051

Hironobu Nakano, Ikuo Yamamoto, Ikuo Yoshida, “Signal processing and image processing using wavelets”, Kyoritsu Shuppan, August 1999, p. 104-110

  As described in Patent Document 1 and Patent Document 2, when the noise removal processing method described in Non-Patent Document 1 is implemented as a circuit that performs stream processing, a circuit in which various memories are distributed and arranged is obtained. . FIG. 18 is an explanatory diagram illustrating an example of a noise removal processing circuit that performs multi-resolution analysis in which wavelet transformation is performed three times and wavelet degeneration. Patent Documents 1 and 2 disclose solution means for reducing these necessary memory amounts, but the noise removal processing circuit illustrated in FIG. 18 does not completely eliminate these solution means. Absent.

  However, the methods described in Patent Document 1 and Patent Document 2 have a plurality of problems. The first is that the total amount of memory required is large. The reason is that since the pixel value with decimal precision frequency-converted by the wavelet transform is stored in the memory, the amount of data per pixel increases compared to the input pixel value with integer precision.

  Second, a large area is required because the memory is distributed. This is because the overhead of the area related to one memory is accumulated as compared with the case where a single memory is arranged.

  The third is that the number of control circuits that read and write memory is large, and each has a different configuration, which makes control complicated. This is because many various means in the circuit exchange data via various memories.

  Accordingly, an object of the present invention is to provide a noise removal system, a noise removal processing circuit, and a noise removal processing method that can remove noise contained in a signal while suppressing an increase in the memory to be used.

  The noise elimination processing system according to the present invention stores a plurality of lines of pixels received in units of lines, reads a plurality of lines of pixels stored in the storage means, and reads each line included in the plurality of lines at the same timing. First wavelet transform means for performing wavelet transform to generate a plurality of first high-frequency components and a plurality of first low-frequency components; and a wavelet transform at the same timing for each of the plurality of first low-frequency components. And a second wavelet transform unit that generates a second high frequency component and a second low frequency component, a first wavelet degeneration unit that degenerates a noise component from the plurality of first high frequency components, A second wavelet degeneration means for degenerating a noise component from the two high frequency components, a second high frequency component with the noise component degenerated, and a second A first inverse wavelet transform unit that performs an inverse wavelet transform based on the low-frequency component and generates and outputs a first inverse-transform signal, and a plurality of first high-frequency components in which noise components are degenerated And a second inverse wavelet transform unit that performs inverse wavelet transform based on the first inverse transform signal and generates and outputs a second inverse transform signal.

  The noise removal processing circuit according to the present invention reads a plurality of lines of pixels stored in the storage circuit and stores a plurality of lines of pixels received in units of lines, and reads each line included in the plurality of lines at the same timing. A first wavelet transform circuit that performs wavelet transform and generates a plurality of first high-frequency components and a plurality of first low-frequency components, and a wavelet transform at the same timing for each of the plurality of first low-frequency components And a second wavelet transform circuit that generates a second high-frequency component and a second low-frequency component, a first wavelet degeneration circuit that degenerates a noise component from the plurality of first high-frequency components, A second wavelet degeneration circuit that degenerates a noise component from the two high-frequency components, a second high-frequency component in which the noise component is degenerated, and a second A first inverse wavelet transform circuit that performs inverse wavelet transform on the basis of the frequency component, generates and outputs a first inverse transform signal, and a plurality of first high-frequency components in which noise components are degenerated and the first And a second inverse wavelet transform circuit that performs inverse wavelet transform based on the inverse transform signal and generates and outputs a second inverse transform signal.

  The noise removal processing method according to the present invention accumulates a plurality of lines of pixels received in units of lines, reads the accumulated pixels of the plurality of lines, and performs wavelet transform at the same timing on each line included in the plurality of lines. To generate a plurality of first high-frequency components and a plurality of first low-frequency components, and perform a wavelet transform on each of the plurality of first low-frequency components at the same timing, thereby obtaining a second high-frequency component. And the second low-frequency component, the noise component is degenerated from the plurality of first high-frequency components, the noise component is degenerated from the second high-frequency component, and the second high-frequency component in which the noise component is degenerated By performing inverse wavelet transform based on the second low-frequency component, a first inverse transform signal is generated and output, and a plurality of noise components are degenerated. By performing the inverse wavelet transform based on the first high frequency component and a first inverted signal, and outputs to generate a second inverted signal.

  According to the present invention, it is possible to remove noise contained in a signal while suppressing an increase in the memory used.

It is a block diagram which shows one Embodiment of the noise removal processing system by this invention. It is explanatory drawing which shows the structural example of the 1st wavelet transform means. It is explanatory drawing which shows the structural example of the 2nd wavelet transformation means. It is explanatory drawing which shows the structural example of the 1st wavelet reduction means 140. FIG. It is explanatory drawing which shows the structural example of the 2nd wavelet reduction means 150. FIG. It is explanatory drawing which shows the operation example of a noise removal processing system. It is a block diagram which shows one Embodiment of the noise removal processing circuit by this invention. 3 is an explanatory diagram illustrating a configuration example of a first wavelet transform circuit 2020. FIG. FIG. 11 is an explanatory diagram showing a configuration example of a second wavelet transform circuit 2030. FIG. 11 is an explanatory diagram showing a configuration example of a third wavelet transform circuit 2040. It is explanatory drawing which shows the example of wavelet transformation. It is explanatory drawing which shows the other example of wavelet transformation. It is explanatory drawing which shows the operation example of a noise removal processing circuit. It is a block diagram which shows the outline | summary of the noise removal processing system by this invention. It is a block diagram which shows the outline | summary of the noise removal processing circuit by this invention. It is explanatory drawing which shows the state which applied 2D wavelet transform to the 2D image. It is explanatory drawing which shows an image noise removal processing method. It is explanatory drawing which shows the example of a noise removal process circuit.

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

  FIG. 1 is a block diagram showing an embodiment of a noise removal processing system according to the present invention. The unidirectional arrows shown in FIG. 1 simply indicate the direction of information flow, and do not exclude bidirectionality.

  The noise removal processing system 100 illustrated in FIG. 1 includes a storage unit 110, a first wavelet transform unit 120, a second wavelet transform unit 130, a first wavelet reduction unit 140, and a second wavelet reduction unit. 150, a second inverse wavelet transform unit 160, and a first inverse wavelet transform unit 170.

  The storage unit 110 accepts pixel value input in units of lines and accumulates a plurality of predetermined lines. In the first embodiment, a case where the storage unit 110 outputs accumulated data of two lines will be described. The storage unit 110 is realized by a memory or the like.

  The first wavelet transform unit 120 reads a plurality of lines from the storage unit 110. The first wavelet transform unit 120 performs two-dimensional wavelet transform on a plurality of lines at the same timing, and separates the plurality of lines into a plurality of first high frequency components and a plurality of first low frequency components. In addition, the same timing here represents simultaneous or substantially simultaneous.

  The second wavelet transform unit 130 receives the plurality of first low frequency components, performs wavelet transform on the plurality of first low frequency components at the same timing, and performs the second high frequency component and the second low frequency component. Separate into frequency components. Note that the same timing here also represents the same time or almost the same time.

  The first wavelet reduction means 140 receives a plurality of first high frequency components and reduces noise components.

  The second wavelet reduction means 150 receives the second high frequency component and reduces the noise component.

  The second inverse wavelet transform unit 170 receives the second high-frequency component and the second low-frequency component from which noise is degenerated, performs inverse wavelet transform, and generates a second inverse transform signal.

  The first inverse wavelet transform unit 160 receives a plurality of first high-frequency components from which noise is degenerated and a second inverse transform signal, and generates a pixel (first inverse transform signal) from which noise is degenerated. ,Output.

  FIG. 2 is an explanatory diagram showing a configuration example of the first wavelet transform unit 120. The first wavelet transform unit 120 includes a two-dimensional wavelet transform unit 121 and a two-dimensional wavelet transform unit 122 that separate a plurality of read lines into a low-frequency component corresponding to one pixel and a high-frequency component corresponding to three pixels. .

  FIG. 3 is an explanatory diagram showing a configuration example of the second wavelet transform unit 130. The second wavelet transform unit 130 includes a two-dimensional wavelet transform unit 131. The contents of the two-dimensional wavelet transform unit 131 are the same as those of the two-dimensional wavelet transform unit 121 or the two-dimensional wavelet transform unit 122 included in the first wavelet transform unit 120.

  FIG. 4 is an explanatory diagram showing a configuration example of the first wavelet reduction means 140. The first wavelet reduction means 140 includes a noise reduction means 141 and a noise reduction means 142 that receive high frequency components and reduce noise components.

  FIG. 5 is an explanatory diagram showing a configuration example of the second wavelet reduction means 150. Second wavelet reduction means 150 includes noise reduction means 151. The contents of the noise reduction means 151 are the same as those of the noise reduction means 141 or the noise reduction means 142 included in the first wavelet reduction means 140.

  The first inverse wavelet transform unit 170 accepts a low frequency component corresponding to one pixel and a high frequency component corresponding to three pixels, and generates two inverse two-dimensional inverse wavelet transform units (not shown). including. The second inverse wavelet transform unit 160 includes a two-dimensional inverse wavelet transform unit (not shown) similar to the two-dimensional inverse wavelet transform unit included in the first inverse wavelet transform unit 170.

  Each configuration illustrated in FIG. 1 generally operates as follows.

  The storage unit 110 accepts input in units of lines and accumulates a plurality of lines. The first wavelet transform unit 120 reads the pixel values of the accumulated plurality of lines.

  The first wavelet transform unit 120 reads a plurality of lines, and the two-dimensional wavelet transform unit 121 and the two-dimensional wavelet transform unit 122 respectively input the read lines. The two-dimensional wavelet transform unit 121 and the two-dimensional wavelet transform unit 122 operate simultaneously and output the first high-frequency component and the first low-frequency component, respectively. Note that the simultaneous operation includes not only the timing of operating completely simultaneously but also the timing of operating substantially simultaneously within a range that does not affect the processing of the subsequent second wavelet transform unit 130.

  The second wavelet transform unit 130 receives the first low-frequency component output from the two-dimensional wavelet transform unit 121 and the two-dimensional wavelet transform unit 122. The two-dimensional wavelet transform unit 131 separates the input first low frequency component into a second high frequency component and a second low frequency component.

  The first wavelet degeneration means 140 inputs the first high frequency component. The second wavelet degeneration means 150 inputs the second high frequency component. The first wavelet reduction means 140 and the second wavelet reduction means 150 generate a high frequency component in which noise is reduced from the input high frequency component.

  The second inverse wavelet transform unit 170 receives the second low frequency component and the second high frequency component after noise reduction, and generates a second inverse transform signal.

  The first inverse wavelet transform unit 160 receives the second inverse transform signal and the first high frequency component after noise reduction, and generates a first inverse transform signal. The noise removal processing system 100 outputs the generated first inverse transform signal.

  First wavelet transform means 120, second wavelet transform means 130, first wavelet reduction means 140, second wavelet reduction means 150, second inverse wavelet transform means 160, and first inverse The wavelet transform unit 170 may be realized as a part of the noise removal processing circuit.

  Next, the operation of the noise removal processing system of this embodiment will be described. FIG. 6 is an explanatory diagram illustrating an operation example of the noise removal processing system of the present embodiment. When pixels are input to the storage unit 110 in units of lines, the storage unit 110 accumulates pixel values of a plurality of lines (step A1).

  Next, the first wavelet transform unit 120 reads a plurality of lines simultaneously from the storage unit 110, performs two two-dimensional wavelet transforms in parallel, and separates them into a first low-frequency component and a first high-frequency component ( Step A2).

  The first wavelet transform unit 120 immediately inputs the generated first low-frequency component to the second wavelet transform unit 130. The second wavelet transform unit 130 performs the two-dimensional wavelet transform again, and separates the input first low-frequency component into a second low-frequency component and a second high-frequency component (step A3).

  The second wavelet reduction means 150 performs noise component reduction processing on the input second high frequency component, and outputs the second high frequency component after the noise removal processing (step A4). The second inverse wavelet transform means 170 receives the second high frequency component and the second low frequency component after the noise removal processing, and performs the two-dimensional inverse wavelet transform. The second inverse wavelet transform unit 170 outputs the generated second inverse transform signal (step A5).

  On the other hand, the first wavelet reduction means 140 performs noise component reduction processing on the input first high frequency component, and outputs the first high frequency component after the noise removal processing (step A6). Finally, the first inverse wavelet transform unit 160 inputs the first high-frequency component and the second inverse transform signal after the noise removal processing, and performs the two-dimensional inverse wavelet transform. The first inverse wavelet transform unit 160 outputs the generated first inverse transform signal. As a result, the first inverse transform signal is output as the processing result of the noise removal processing system 100 (step A7).

  Next, the effect of this embodiment will be described. As described above, in the present embodiment, the storage unit 110 accumulates a plurality of lines of received pixels in units of lines, and the first wavelet transform unit 120 reads the accumulated pixels of the plurality of lines and stores them in the plurality of lines. A plurality of first high frequency components and a plurality of first low frequency components are generated by performing wavelet transform at the same timing on each included line. In addition, the second wavelet transform unit 130 performs the wavelet transform on the plurality of first low frequency components at the same timing, thereby generating the second high frequency component and the second low frequency component. Further, the first wavelet reduction means 140 reduces the noise component from the plurality of first high frequency components, and the second wavelet reduction means 140 reduces the noise component from the second high frequency component. Then, the second inverse wavelet transform unit 170 performs the inverse wavelet transform based on the second high-frequency component and the second low-frequency component in which the noise component is degenerated, thereby obtaining the first inverse-transform signal. The first inverse wavelet transforming unit 170 performs the inverse wavelet transform on the basis of the plurality of first high frequency components and the first inverse transform signal in which the noise components are degenerated. 2 is generated and output. Therefore, noise included in the signal can be removed while suppressing an increase in the memory to be used.

  That is, in the present embodiment, a plurality of lines of input pixels are first accumulated and then a plurality of lines of pixels are supplied to the first wavelet transform unit 120. The first wavelet transform unit 120 simultaneously generates a low-frequency component necessary for the second wavelet transform by performing a plurality of two-dimensional wavelet transforms. Therefore, the first wavelet transform and the second wavelet transform are performed almost simultaneously.

  Further, since the accumulated data is not the data after frequency conversion but an integer pixel value (integer precision data), the amount of accumulated data is reduced as compared with a general method. In addition, since the memory unit 110 can be realized by a single memory without being distributed, the area is reduced by reducing the number of circuits for controlling the memory unit 110. Further, since the number of control circuits is reduced, the control is simplified.

  In the above description, the case where the noise removal processing system performs wavelet transformation twice is exemplified. However, the number of wavelet transforms performed by the noise removal processing system of the present invention is not limited to two, and may be three or more.

In this case, one or more intermediate wavelet transform means (not shown) for generating a plurality of high frequency components and a plurality of low frequency components from the plurality of low frequency components are the first wavelet transform means and the second wavelet transform means. It may be provided between. In this case, as in the noise removal system 100 shown in FIG. 1, one or more intermediate wavelet reduction means (not shown) for reducing noise components from a plurality of high frequency components are connected to each of the intermediate wavelet transform means. Further, one or more intermediate inverse wavelet transform means (not shown) that performs inverse wavelet transform based on a plurality of high-frequency components with reduced noise components and the inverse transform signal, and generates and outputs a new inverse transform signal. ) Is connected to each of the intermediate wavelet degeneration means.
That is, the number of intermediate wavelet transform means, intermediate wavelet degeneration means, and intermediate inverse wavelet transform means is the same.

  The first wavelet transform unit inputs the first low frequency component to the intermediate wavelet transform unit, and the intermediate wavelet transform unit converts the generated plurality of low frequency components into another intermediate wavelet transform unit or the second wavelet transform unit. Input to wavelet transform means. The first inverse wavelet transform unit inputs the generated first inverse transform signal to the intermediate inverse wavelet transform unit, and the intermediate inverse wavelet transform unit converts the generated inverse transform signal into another intermediate inverse wavelet transform unit. Or it inputs into the 2nd inverse wavelet transform means.

  The specific contents of these intermediate wavelet transform means, intermediate wavelet degeneration means, and intermediate inverse wavelet transform means will be described in the configuration of a noise removal processing circuit described later.

  Next, the noise removal processing circuit of the present invention will be described using a specific example. FIG. 7 is a block diagram showing an embodiment of a noise removal processing circuit according to the present invention. The noise removal processing circuit illustrated in FIG. 7 is a noise removal processing circuit using multi-resolution analysis that performs wavelet transformation three times. However, the number of wavelet transformations performed by the noise removal processing circuit of the present invention is not limited to three, and may be two or four or more.

  The noise removal processing circuit 2000 illustrated in FIG. 7 includes a line buffer 2010, a first wavelet transform circuit 2020, a second wavelet transform circuit 2030, a third wavelet transform circuit 2040, and a first wavelet degeneration circuit. 2050, second wavelet reduction circuit 2060, third wavelet reduction circuit 2070, first inverse wavelet transformation circuit 2080, second inverse wavelet transformation circuit 2090, and third inverse wavelet transformation circuit 2100 It has.

  The second wavelet transform circuit 2030, the second wavelet degeneration circuit 2060, and the second inverse wavelet transform circuit 2090 correspond to the above-described intermediate wavelet transform unit, intermediate wavelet degenerate unit, and intermediate inverse wavelet transform unit, respectively. .

  The line buffer 2010 accepts pixel input in units of lines and accumulates a plurality of lines.

  The first wavelet transform circuit 2020 includes four two-dimensional wavelet transform circuits therein. The first wavelet transform circuit 2020 reads a plurality of lines from the line buffer 2010 into a first low-frequency component corresponding to 4 pixels and a first high-frequency component corresponding to 12 pixels (3 pixels × 4 parallel). To separate.

  Second wavelet transform circuit 2030 includes two two-dimensional wavelet transform circuits therein. The second wavelet transform circuit 2030 receives the first low-frequency component and separates it into a second low-frequency component corresponding to two pixels and a second high-frequency component.

  Third wavelet transform circuit 2040 includes one two-dimensional wavelet transform circuit therein. The third wavelet transform circuit 2040 receives the second low frequency component and separates it into a third low frequency component and a third high frequency component.

  The first wavelet reduction circuit 2050 includes four noise reduction circuits inside. Second wavelet reduction circuit 2060 includes two noise reduction circuits. Third wavelet degeneration circuit 2070 includes one noise degeneration circuit.

  The first inverse wavelet transform circuit 2080 includes four two-dimensional inverse wavelet transform circuits therein. The second inverse wavelet transform circuit 2090 includes two two-dimensional inverse wavelet transform circuits therein. The third inverse wavelet transform circuit 2100 includes one two-dimensional inverse wavelet transform circuit therein.

  The noise removal processing circuit 2000 is realized by a dedicated circuit such as an ASIC (Application Specified Integral Circuit). The noise removal processing circuit 2000 may be realized by a rewritable circuit device such as an FPGA (Field Programmable Gate Array).

  The line buffer 2010 is realized by, for example, a volatile memory element such as a static random access memory (SRAM) or a dynamic random access memory (DRAM). In addition, the line buffer 2010 includes an MRAM (Magnetic Resistive RAM), a PRAM (Phase Change RAM), a ReRAM (Resistive RAM, a resistance change memory), an STT-RAM (Spin Transfer Torque RAM). , Spin injection memory) or the like.

  The line buffer 2010 may be realized by a hard disk (HD), a solid state drive (SSD), a flash memory, or the like, or may be realized by a disk medium such as an FD, a DVD, or a CD. However, the example given above shows an example of the line buffer 2010 and does not limit the application. The line buffer 2010 can be realized as long as it can record information.

  FIG. 8 is an explanatory diagram illustrating a configuration example of the first wavelet transform circuit 2020. The first wavelet transform circuit 2020 includes a two-dimensional wavelet transform circuit 2021, a two-dimensional wavelet transform circuit 2022, a two-dimensional wavelet transform circuit 2023, and a two-dimensional wavelet transform circuit 2024.

  FIG. 9 is an explanatory diagram showing a configuration example of the second wavelet transform circuit 2030. The second wavelet transform circuit 2030 includes a two-dimensional wavelet transform circuit 2031 and a two-dimensional wavelet transform circuit 2032.

  FIG. 10 is an explanatory diagram showing a configuration example of the third wavelet transform circuit 2040. The third wavelet transform circuit 2040 includes a two-dimensional wavelet transform circuit 2041 and a two-dimensional wavelet transform circuit.

  The first wavelet degeneration circuit 2050 includes four noise degeneration circuits (not shown), the second wavelet degeneration circuit 2060 includes two noise reduction circuits (not shown), and a third wavelet. Degeneration circuit 2070 includes one noise degeneration circuit (not shown).

  The first inverse wavelet transform circuit 2080 includes four two-dimensional inverse wavelet transform circuits (not shown), and the second inverse wavelet transform circuit 2090 includes two two-dimensional inverse wavelet transform circuits (not shown). The third inverse wavelet transform circuit 2100 includes one two-dimensional inverse wavelet transform circuit (not shown).

  The noise removal processing circuit 2000 including the above configuration generally operates as follows.

  FIG. 11 is an explanatory diagram showing an example of wavelet transform in this example. In the example illustrated in FIG. 11, the third wavelet transform is performed from the frequency component after the first wavelet transform. In addition, L1 to L3 of the pixel illustrated in FIG. 11 indicate first to third low frequency components, and H1 to H3 of the pixel respectively indicate first to third high frequency components.

  The first wavelet transform circuit 2020, the second wavelet transform circuit 2030, and the third wavelet transform circuit 2040 each include a two-dimensional wavelet transform circuit. Here, the two-dimensional wavelet transform is realized by performing a discrete wavelet transform in the vertical direction and a discrete wavelet transform in the horizontal direction, respectively. The wavelet transform includes a basic function that exists locally in space. This is called a mother wavelet, and the accuracy differs depending on the type of the mother wavelet.

  For example, in the simplest Haar Wavelet, a low frequency component of 1 pixel and a low frequency component of 3 pixels are generated from a 2 × 2 input pixel. Hereinafter, this specific example will be described using this Haar Wavelet.

  As illustrated in FIG. 11C, in order to perform the third wavelet transform, a second low-frequency component corresponding to four pixels is necessary. In order to generate the second low-frequency component of 4 pixels, the first low-frequency component of 16 pixels corresponding to 4 × 4 pixels is necessary as illustrated in FIG. In order to generate the first low-frequency component of 16 pixels, it is necessary to input 8 × 8 original pixels as illustrated in FIG. 11A to the first wavelet transform.

  From the above, in order to perform the first to third wavelet transforms almost simultaneously, it is sufficient that the original input pixels for 8 lines are stored in the line buffer 2010. The number of accumulated lines varies depending on the type of mother wavelet used. Note that the specific example shown in FIG. 11 is an example, and does not limit the number of lines to be accumulated.

  FIG. 12 is an explanatory diagram showing another example of the wavelet transform in this specific example. In the example shown in FIG. 12, the first to third wavelet transforms are performed. Specifically, first, the first wavelet transform includes four two-dimensional wavelet transforms. Thereby, the low frequency component produced | generated simultaneously is for 4 pixels in a perpendicular direction. When the second first wavelet transform is performed, a first low-frequency component corresponding to 2 × 2 pixels necessary for the second wavelet transform is generated. Further, when the process proceeds and the first first wavelet transform is performed, the second second wavelet transform is performed. As a result, a second low-frequency component corresponding to 2 × 2 pixels necessary for the third wavelet transform is generated, and the third wavelet transform is performed.

  As described above, in this specific example, the first to third wavelet transforms are not performed at the same time. However, since no delay occurs in line units, a memory for accumulating low frequency components and a memory for accumulating high frequency components are not required. For this reason, it is possible to eliminate the memory for storing the intermediate data.

In this specific example, the most efficient configuration example for achieving the object of the invention is shown. Generalizing this specific example, when the number of times of performing the wavelet transform is ^N , 2 ^(N−1) two-dimensional wavelet transform circuits are included in order to perform the first wavelet transform.

  The number of two-dimensional wavelet transform circuits included in each wavelet transform circuit may be equal to or greater than the number of parallel in the vertical direction shown in this specific example. For example, in the first wavelet transformation in this specific example, the noise removal processing circuit 2000 arranges four two-dimensional wavelet transformation circuits in the vertical direction and further arranges 16 two-dimensional wavelets in which four circuits are arranged in the horizontal direction. A configuration including conversion may be used.

  In the case of this configuration, the second low-frequency component corresponding to 2 × 2 pixels for performing the third wavelet transform can be generated simultaneously by performing the first wavelet transform only once. Even when any of the above configurations is adopted, the memory for storing the low frequency components in line units, that is, the memory for storing the intermediate data is eliminated, and the object of the present invention is achieved.

  Next, an operation example of the noise removal processing circuit according to the present invention will be described. FIG. 13 is an explanatory diagram showing an operation example of the noise removal processing circuit according to the present invention. When the current pixel is input to the line buffer 2010, the line buffer 2010 accumulates the pixel for a plurality of lines (step B1). In this specific example in which the wavelet transform is performed three times, when the mother wavelet is Haar Wavelet, the number of accumulated lines is eight.

  The first wavelet transform circuit 2020 reads out pixels of a plurality of lines simultaneously after a predetermined number of lines are accumulated in the line buffer 2010. The first wavelet transform circuit 2020 separates the first low-frequency component and the first high-frequency component corresponding to four pixels by the four two-dimensional wavelet transform circuits operating in parallel (step B2).

  The second wavelet transform circuit 2030 immediately inputs the first low frequency component generated in step B2, and separates it into a second low frequency component and a second high frequency component (step B3). The third wavelet transform circuit 2040 immediately inputs the second low-frequency component generated in step B3 and separates it into a third low-frequency component and a third high-frequency component (step B4).

  The third wavelet degeneration circuit 2070 receives the third high-frequency component, and generates a third high-frequency component with noise reduced (step B5). The third inverse wavelet transform circuit 2100 receives the third high-frequency component and the third low-frequency component from which noise is degenerated, and generates a third inverse transform signal (step B6).

  On the other hand, the second wavelet degeneration circuit 2060 receives the second high-frequency component and generates a second high-frequency component with noise reduced (step B7). The second inverse wavelet transform circuit 2090 receives the second high-frequency component from which noise has been degenerated and the third inverse transform signal, and generates a second inverse transform signal (step B8).

  Further, the first wavelet degeneration circuit 2050 receives the first high frequency component and generates a first high frequency component with noise reduced (step B9). The first inverse wavelet transform circuit 2080 receives the first high frequency component from which noise is degenerated and the second inverse transform signal, and generates a first inverse transform signal. The noise removal processing circuit 2000 outputs a first inverse conversion signal as a processing result (step B10).

  In this specific example, the case where the first wavelet degeneration circuit includes four noise degeneration circuits has been described. However, there is no particular limitation on the included noise degeneration circuit.

  Further, as illustrated in FIG. 13, the second and third wavelet transforms and the second and third wavelet degeneration processes or the second and third wavelet degeneration processes or the first and second wavelet transforms from the execution of step B9 to the execution of step B10. 2. The third inverse wavelet transform needs to be executed and finished. In other words, there is no problem if Step B9 is completed before Step B8 is completed. Further, the parallelism of the noise reduction circuit included in the first wavelet reduction circuit 2050 is not particularly limited.

  In this specific example, Haar Wavelet is assumed. On the other hand, in the discrete wavelet transform, there is also a method of generating a low frequency component corresponding to one pixel from seven pixels or nine pixels. When using such a wavelet transform, it is possible to obtain the same effect as in the above specific example by further increasing the degree of parallelism.

Specifically, when N wavelet transforms are performed, a parallelism of 2 ^(N−1) +6 is required for a 7 pixel wavelet transform, and 2 for a 9 pixel wavelet transform. ^(N-1) +8 parallelism is required. In general, when the wavelet transform of K pixels is performed, the degree of parallelism shown in the following formula 1 is required.

  Next, the outline of the present invention will be described. FIG. 14 is a block diagram showing an outline of a noise removal processing system according to the present invention. A noise removal processing system 80 illustrated in FIG. 14 reads a storage unit 81 (for example, the storage unit 110) that accumulates a plurality of pixels received in units of lines, and reads a plurality of lines of pixels stored in the storage unit 81. A first wavelet transform unit 82 (for example, a first wavelet transform unit 82) that performs wavelet transform on each line included in the plurality of lines at the same timing to generate a plurality of first high-frequency components and a plurality of first low-frequency components. Wavelet transform means 120) and second wavelet transform means for generating a second high-frequency component and a second low-frequency component by performing wavelet transform on the plurality of first low-frequency components at the same timing. 83 (for example, the second wavelet transform unit 130) and a first for degenerating a noise component from the plurality of first high frequency components Wavelet reduction means 84 (for example, first wavelet reduction means 140), second wavelet reduction means 85 (for example, second wavelet reduction means 150) for reducing the noise component from the second high frequency component, noise component The first inverse wavelet transform means 86 (for example, for generating and outputting the first inverse transform signal by performing inverse wavelet transform based on the second high frequency component and the second low frequency component in which is degenerated Based on the second inverse wavelet transform means 170), the plurality of first high frequency components with the noise components degenerated and the first inverse transform signal, the inverse wavelet transform is performed to generate the second inverse transform signal. Second inverse wavelet transforming means 87 (for example, first inverse wavelet transforming means 160) for output.

  With such a configuration, noise included in the signal can be removed while suppressing an increase in the memory to be used.

  The noise removal processing system 80 is provided between the first wavelet transform unit 82 and the second wavelet transform unit 83, and generates a plurality of high frequency components and a plurality of low frequency components from the plurality of low frequency components. One or more intermediate wavelet transform means (for example, the second wavelet transform circuit 2030) and one or more intermediate wavelet degeneration means (for example, the second wavelet transform circuit 2030) that are connected to each of the intermediate wavelet transform means and degenerate a noise component from a plurality of high-frequency components. The second wavelet degeneration circuit 2060) and the intermediate wavelet degeneration means are connected to each other, and the inverse wavelet transform is performed based on the plurality of high frequency components with the noise components degenerated and the inverse transform signal, and a new inverse transform signal is obtained. One or more intermediate inverse wavelet transform means to generate and output (eg, second inverse way) Wavelet transform circuit) and may comprise. The first wavelet transform unit 82 inputs the first low-frequency component to the intermediate wavelet transform unit, and the intermediate wavelet transform unit converts the generated plurality of low-frequency components into another intermediate wavelet transform unit or the second wavelet transform unit. The first inverse wavelet transform unit 82 inputs the generated first inverse transform signal to the intermediate inverse wavelet transform unit, and the intermediate inverse wavelet transform unit outputs the generated inverse transform signal. 87 may be input to other intermediate inverse wavelet transform means or second inverse wavelet transform means.

  With such a configuration, it becomes possible to process more line inputs collectively.

  Each wavelet transform unit may include two-dimensional wavelet transform units (for example, two-dimensional wavelet transform units 121, 122, and 131) having different parallel degrees.

  At this time, the parallelism of the two-dimensional wavelet transform unit included in the wavelet transform unit that performs wavelet transform using the generated low-frequency component is the two-dimensional wavelet included in the wavelet transform unit that is the transmission source of the low-frequency component. Less than parallelism of conversion means.

  Specifically, the wavelet transforming means for transmitting the low frequency component to the other wave red transforming means is a two-dimensional two-dimensional wavelet transforming means compared with the two-dimensional wavelet transforming means included in the destination wavelet transforming means. May be included.

Further, when the number of times of performing the wavelet transform is N, the first wavelet transform unit generates 2 ^(N−1) first low-frequency components from a plurality of lines of pixels, and each wavelet The converting means may perform the wavelet transform at substantially the same timing (simultaneously or almost simultaneously).

  FIG. 15 is a block diagram showing an outline of a noise removal processing circuit according to the present invention. The noise removal processing circuit 90 illustrated in FIG. 15 reads a storage circuit 91 (for example, a line buffer 2010) that stores a plurality of lines of pixels received in units of lines, and reads a plurality of lines of pixels stored in the storage circuit 91. A first wavelet transform circuit 92 (for example, a first wavelet transform circuit 92) that performs wavelet transform on each line included in the plurality of lines at the same timing to generate a plurality of first high-frequency components and a plurality of first low-frequency components. Wavelet transform circuit 2020) and a second wavelet transform circuit that performs a wavelet transform on the plurality of first low-frequency components at the same timing to generate a second high-frequency component and a second low-frequency component. 93 (for example, the third wavelet transform circuit 2040) and degenerate noise components from the plurality of first high frequency components A first wavelet degeneration circuit 94 (for example, a first wavelet degeneration circuit 2050), and a second wavelet degeneration circuit 95 (for example, a third wavelet degeneration circuit 2070) that degenerates a noise component from the second high-frequency component; A first inverse wavelet transform circuit 96 that performs inverse wavelet transform based on the second high frequency component and the second low frequency component in which the noise component is degenerated, and generates and outputs a first inverse transform signal. (For example, the first inverse wavelet transform circuit 2100), the inverse wavelet transform is performed based on the plurality of first high-frequency components with the noise components degenerated and the first inverse transform signal, and the second inverse transform is performed. A second inverse wavelet transform circuit 97 (for example, a first inverse wavelet transform circuit 2080) that generates and outputs a signal. .

  Even with such a configuration, it is possible to remove noise contained in the signal while suppressing an increase in the memory used.

  The noise removal processing circuit 90 is provided between the first wavelet transform circuit 92 and the second wavelet transform circuit 93, and generates a plurality of high frequency components and a plurality of low frequency components from the plurality of low frequency components. One or more intermediate wavelet transform circuits (for example, the second wavelet transform circuit 2030) and one or more intermediate wavelet degeneration circuits (for example, the second wavelet transform circuit 2030) that are connected to each of the intermediate wavelet transform circuits and degenerate noise components from a plurality of high frequency components (for example, A second wavelet degeneration circuit 2060) and a plurality of high-frequency components with noise components degenerated and an inverse transform signal connected to each of the intermediate wavelet degeneration circuits, and an inverse wavelet transform is performed to generate a new inverse transform signal. One or more intermediate inverse wavelet transform circuits (eg, the second inverse wavelet) that are generated and output. DOO conversion circuit 2090) and may comprise. The first wavelet transform circuit 92 inputs the first low-frequency component to the intermediate wavelet transform circuit, and the intermediate wavelet transform circuit converts the generated plurality of low-frequency components into another intermediate wavelet transform circuit or the second wavelet transform circuit. The first inverse wavelet transform circuit inputs the generated first inverse transform signal to the intermediate inverse wavelet transform circuit, and the intermediate inverse wavelet transform circuit receives the generated inverse transform signal as follows: It may be input to another intermediate inverse wavelet transform circuit or second inverse wavelet transform circuit 97.

  A part or all of the above-described embodiment can be described as in the following supplementary notes, but is not limited thereto.

(Appendix 1) Storage means for accumulating a plurality of lines of pixels received in units of lines, and reading the pixels of a plurality of lines stored in the storage means, and performing wavelet transform at the same timing for each line included in the plurality of lines First wavelet transforming means for generating a plurality of first high-frequency components and a plurality of first low-frequency components, and performing a wavelet transform on the plurality of first low-frequency components at the same timing. , Second wavelet transform means for generating a second high frequency component and a second low frequency component, first wavelet degeneration means for degenerating noise components from the plurality of first high frequency components, and the second Second wavelet degeneration means for degenerating a noise component from the high frequency component of the second, the second high frequency component with the noise component degenerated and the second A first inverse wavelet transform unit that performs an inverse wavelet transform based on the frequency component to generate and output a first inverse transform signal; the plurality of first high frequency components in which a noise component is degenerated; A noise removal processing system comprising: a second inverse wavelet transform unit that performs an inverse wavelet transform based on the first inverse transform signal and generates and outputs a second inverse transform signal.

(Appendix 2) One or more intermediate wavelet transforms that are provided between the first wavelet transform unit and the second wavelet transform unit and generate a plurality of high frequency components and a plurality of low frequency components from a plurality of low frequency components And at least one intermediate wavelet reduction means connected to each of the intermediate wavelet transform means for reducing noise components from the plurality of high frequency components, and a plurality of noise components reduced by being connected to each of the intermediate wavelet reduction means One or more intermediate inverse wavelet transforming means for generating and outputting a new inverse transform signal based on the high-frequency component and the inverse transform signal, and outputting a new inverse transform signal. The first wavelet transform means comprises: 1 low frequency component is input to the intermediate wavelet transform unit, and the intermediate wavelet transform unit generates A number of low frequency components are input to other intermediate wavelet transform means or second wavelet transform means, and the first inverse wavelet transform means inputs the generated first inverse transform signal to the intermediate inverse wavelet transform means. And the intermediate inverse wavelet transform unit inputs the generated inverse transform signal to another intermediate inverse wavelet transform unit or second inverse wavelet transform unit.

(Supplementary note 3) The noise removal processing system according to supplementary note 1 or supplementary note 2, wherein each wavelet transform unit includes two-dimensional wavelet transform units having different parallel degrees.

(Supplementary Note 4) The degree of parallelism of the two-dimensional wavelet transform unit included in the wavelet transform unit that performs wavelet transform using the generated low-frequency component is the two-dimensionality included in the wavelet transform unit that is the transmission source of the low-frequency component The noise removal processing system according to supplementary note 3, wherein the parallelism of the wavelet transform means is less.

(Additional remark 5) The wavelet transformation means which transmits a low-frequency component to other wave red transformation means is a two-dimensional two-dimensional wavelet transformation means compared with the two-dimensional wavelet transformation means included in the destination wavelet transformation means. The noise removal processing system according to Appendix 3 or Appendix 4.

(Supplementary Note 6) When the number of times of performing wavelet transform is N, the first wavelet transform unit generates 2 ^(N−1) first low-frequency components from a plurality of lines of pixels, The wavelet transform means is the noise removal processing system according to any one of appendix 1 to appendix 5, in which the wavelet transform is performed at substantially the same timing.

(Supplementary note 7) A storage circuit for storing a plurality of lines of pixels received in units of lines, and a plurality of lines of pixels stored in the storage circuit are read, and wavelet transform is performed on each line included in the plurality of lines at the same timing. A first wavelet transform circuit that generates a plurality of first high-frequency components and a plurality of first low-frequency components, and a wavelet transform is performed at the same timing on each of the plurality of first low-frequency components. , A second wavelet transform circuit that generates a second high frequency component and a second low frequency component, a first wavelet degeneration circuit that degenerates a noise component from the plurality of first high frequency components, and the second A second wavelet degeneration circuit that degenerates a noise component from the high-frequency component of the second, a second high-frequency component in which the noise component is degenerated, and the second A first inverse wavelet transform circuit that performs inverse wavelet transform based on the frequency component, generates and outputs a first inverse transform signal, and the plurality of first high frequency components in which noise components are degenerated and A noise removal processing circuit comprising: a second inverse wavelet transform circuit that performs inverse wavelet transform based on the first inverse transform signal and generates and outputs a second inverse transform signal.

(Appendix 8) One or more intermediate wavelet transforms that are provided between the first wavelet transform circuit and the second wavelet transform circuit and generate a plurality of high frequency components and a plurality of low frequency components from a plurality of low frequency components A circuit, one or more intermediate wavelet degeneration circuits that are connected to each of the intermediate wavelet transform circuits and degenerate a noise component from the plurality of high-frequency components, and a plurality of noise components that are connected to each of the intermediate wavelet degeneration circuits The inverse wavelet transform is performed based on the high-frequency component and the inverse transform signal, and includes one or more intermediate inverse wavelet transform circuits that generate and output a new inverse transform signal. The first wavelet transform circuit includes: 1 low frequency component is input to the intermediate wavelet transform circuit, and the intermediate wavelet transform circuit generates A number of low frequency components are input to another intermediate wavelet transform circuit or a second wavelet transform circuit, and the first inverse wavelet transform circuit inputs the generated first inverse transform signal to the intermediate inverse wavelet transform circuit. The noise removal processing circuit according to appendix 7, wherein the intermediate inverse wavelet transform circuit inputs the generated inverse transform signal to another intermediate inverse wavelet transform circuit or a second inverse wavelet transform circuit.

(Supplementary note 9) The noise removal processing circuit according to supplementary note 7 or supplementary note 8, wherein each of the wavelet transform circuits includes a two-dimensional wavelet transform circuit having a different degree of parallelism.

(Supplementary Note 10) The degree of parallelism of the two-dimensional wavelet transformation circuit included in the wavelet transformation circuit that performs wavelet transformation using the generated low-frequency component is the two-dimensionality included in the wavelet transformation circuit that is the transmission source of the low-frequency component. The noise removal processing circuit according to appendix 9, wherein the parallelism of the wavelet transform circuit is less.

(Additional remark 11) The wavelet transformation circuit which transmits a low frequency component to another wave red transformation circuit has a two-dimensional two-dimensional wavelet transformation circuit compared with the two-dimensional wavelet transformation circuit contained in the wavelet transformation circuit of a transmission destination. The noise removal processing circuit according to appendix 9 or appendix 10.

(Additional remark 12) When the frequency | count which performs wavelet transformation is set to N, a 1st wavelet transformation circuit produces | generates 2 ^(N-1) 1st low frequency components from the pixel of several lines, The wavelet transform circuit according to any one of appendix 7 to appendix 11, wherein the wavelet transform is performed at substantially the same timing.

(Supplementary Note 13) A plurality of lines are accumulated by accumulating a plurality of pixels received in line units, reading the accumulated pixels of the plurality of lines, and performing wavelet transform on each line included in the plurality of lines at the same timing. 1 high frequency component and a plurality of first low frequency components are generated, and wavelet transform is performed on each of the plurality of first low frequency components at the same timing, whereby a second high frequency component and a second low frequency component are generated. A low-frequency component, a noise component is degenerated from the plurality of first high-frequency components, a noise component is degenerated from the second high-frequency component, and the second high-frequency component in which the noise component is degenerated and the By performing inverse wavelet transform based on the second low frequency component, the first inverse transformed signal is generated and output, and the noise components are degenerated. By performing the inverse wavelet transform based on the first of the high-frequency component first inverted signal, noise removal processing method characterized by generating and outputting a second inverted signal.

(Additional remark 14) It is a process implemented between the production | generation process of a 1st low frequency component, and the production | generation process of a 2nd low frequency component, Comprising: A several high frequency component and several low frequency from several low frequency components One or more intermediate wavelet transform processes for generating a frequency component, one or more intermediate wavelet degeneration processes for degenerating a noise component from the plurality of high frequency components, a plurality of high frequency components in which the noise component is degenerated, and an inverse transform signal And performing one or more intermediate inverse wavelet transform processes for generating and outputting a new inverse transform signal, and in the intermediate wavelet transform process, a plurality of the inverse wavelet transform processes are performed based on the first low-frequency component. Process of generating a plurality of high frequency components and a plurality of low frequency components, and generating a plurality of high frequency components and a plurality of low frequency components based on the generated plurality of low frequency components at least once Supplementary note 13 wherein the intermediate inverse wavelet transform process generates a new inverse transform signal based on the first inverse transform signal, and generates a new inverse transform signal based on the generated inverse transform signal at least once. Noise removal processing method.

  INDUSTRIAL APPLICABILITY The present invention is suitably applied to uses such as removing noise that deteriorates visibility in systems that require visibility, such as surveillance systems and surveillance cameras. The present invention is also suitably applied to the use of removing noise components that deteriorate the compression efficiency in still image and moving image compression processing and improving the compression efficiency.

  Although the present invention has been described with reference to the embodiments and examples, the present invention is not limited to the above embodiments and examples. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  This application claims the priority on the basis of the JP Patent application 2017-118410 for which it applied on June 16, 2017, and takes in those the indications of all here.

DESCRIPTION OF SYMBOLS 100 Noise removal processing system 110 Memory | storage means 120 1st wavelet transformation means 121,122,131,141,142,151 Two-dimensional wavelet transformation means 130 2nd wavelet transformation means 140 1st wavelet reduction means 150 2nd wavelet Degeneration unit 160 First inverse wavelet transform unit 170 Second inverse wavelet transform unit 2000 Noise removal processing circuit 2010 Line buffer 2020 First wavelet transform circuit 2021, 2022, 2023, 2024 Two-dimensional wavelet transform circuit 2030 Second wavelet Transform circuit 2031, 2032, 2041 Two-dimensional wavelet transform circuit 2040 Third wavelet transform circuit 2050 First wavelet degeneration circuit 206 0 second wavelet degeneration circuit 2070 third wavelet degeneration circuit 2080 first inverse wavelet transform circuit 2090 second inverse wavelet transform circuit 2100 third inverse wavelet transform circuit

Claims (14)

Storage means for accumulating a plurality of lines of pixels received in line units;
The pixels of the plurality of lines stored in the storage unit are read, wavelet transform is performed on each line included in the plurality of lines at the same timing, and a plurality of first high frequency components and a plurality of first low frequency components are First wavelet transform means for generating
A second wavelet transform unit that performs wavelet transform on the plurality of first low-frequency components at the same timing, and generates a second high-frequency component and a second low-frequency component;
First wavelet degeneration means for degenerating a noise component from the plurality of first high frequency components;
Second wavelet degeneration means for degenerating a noise component from the second high frequency component;
First inverse wavelet transform means for performing inverse wavelet transform based on the second high frequency component and the second low frequency component in which noise components are degenerated, and generating and outputting a first inverse transform signal When,
A second inverse wavelet that performs inverse wavelet transform based on the plurality of first high-frequency components degenerated with noise components and the first inverse transform signal, and generates and outputs a second inverse transform signal A noise removal processing system comprising a conversion means. One or more intermediate wavelet transforming means provided between the first wavelet transforming means and the second wavelet transforming means for generating a plurality of high frequency components and a plurality of low frequency components from a plurality of low frequency components;
One or more intermediate wavelet degeneration means connected to each of the intermediate wavelet transform means for degenerating noise components from the plurality of high frequency components;
One or more intermediates connected to each of the intermediate wavelet reduction means, performing inverse wavelet transformation based on a plurality of high frequency components with reduced noise components and the inverse transformation signal, and generating and outputting a new inverse transformation signal An inverse wavelet transform means,
The first wavelet transforming means inputs the first low frequency component to the intermediate wavelet transforming means,
The intermediate wavelet transform unit inputs the plurality of generated low frequency components to another intermediate wavelet transform unit or a second wavelet transform unit,
The first inverse wavelet transform means inputs the generated first inverse transform signal to the intermediate inverse wavelet transform means,
The noise removal processing system according to claim 1, wherein the intermediate inverse wavelet transform unit inputs the generated inverse transform signal to another intermediate inverse wavelet transform unit or a second inverse wavelet transform unit. The noise removal processing system according to claim 1, wherein each wavelet transform unit includes a two-dimensional wavelet transform unit having a different degree of parallelism. The degree of parallelism of the two-dimensional wavelet transform unit included in the wavelet transform unit that performs wavelet transform using the generated low frequency component is the same as that of the two-dimensional wavelet transform unit included in the wavelet transform unit that is the transmission source of the low frequency component. The noise removal processing system according to claim 3, wherein the degree of parallelism is less. The wavelet transform means for transmitting a low frequency component to another wave red transform means includes a two-dimensional two-dimensional wavelet transform means as compared with the two-dimensional wavelet transform means included in the destination wavelet transform means. Or the noise removal processing system of Claim 4. When the number of times of performing the wavelet transform is N, the first wavelet transform means generates 2 ^(N−1) first low-frequency components from a plurality of lines of pixels,
The noise removal processing system according to any one of claims 1 to 5, wherein each wavelet transform unit performs wavelet transform at substantially the same timing. A storage circuit for storing a plurality of lines of pixels received in units of lines;
The pixels of the plurality of lines stored in the storage circuit are read, wavelet transform is performed at the same timing on each line included in the plurality of lines, and a plurality of first high frequency components and a plurality of first low frequency components are A first wavelet transform circuit for generating
A second wavelet transform circuit that performs wavelet transform on the plurality of first low-frequency components at the same timing, and generates a second high-frequency component and a second low-frequency component;
A first wavelet degeneration circuit that degenerates a noise component from the plurality of first high-frequency components;
A second wavelet degeneration circuit that degenerates a noise component from the second high-frequency component;
A first inverse wavelet transform circuit that performs inverse wavelet transform based on the second high-frequency component and the second low-frequency component in which noise components are degenerated, and generates and outputs a first inverse transform signal When,
A second inverse wavelet that performs inverse wavelet transform based on the plurality of first high-frequency components degenerated with noise components and the first inverse transform signal, and generates and outputs a second inverse transform signal A noise removal processing circuit comprising a conversion circuit. One or more intermediate wavelet transform circuits provided between the first wavelet transform circuit and the second wavelet transform circuit and generating a plurality of high frequency components and a plurality of low frequency components from the plurality of low frequency components;
One or more intermediate wavelet degeneration circuits connected to each of the intermediate wavelet transform circuits and degenerating noise components from the plurality of high frequency components;
One or more intermediates connected to each of the intermediate wavelet degeneration circuits, performing inverse wavelet transform on the basis of a plurality of high frequency components with noise components degenerated and inverse transform signals, and generating and outputting new inverse transform signals An inverse wavelet transform circuit,
The first wavelet transform circuit inputs a first low frequency component to the intermediate wavelet transform circuit,
The intermediate wavelet transform circuit inputs the generated plurality of low frequency components to another intermediate wavelet transform circuit or a second wavelet transform circuit,
The first inverse wavelet transform circuit inputs the generated first inverse transform signal to the intermediate inverse wavelet transform circuit,
The noise removal processing circuit according to claim 7, wherein the intermediate inverse wavelet transform circuit inputs the generated inverse transform signal to another intermediate inverse wavelet transform circuit or a second inverse wavelet transform circuit. The noise removal processing circuit according to claim 7, wherein each wavelet transform circuit includes a two-dimensional wavelet transform circuit having a different degree of parallelism. The degree of parallelism of the two-dimensional wavelet transformation circuit included in the wavelet transformation circuit that performs wavelet transformation using the generated low-frequency component is the same as that of the two-dimensional wavelet transformation circuit included in the wavelet transformation circuit that is the transmission source of the low-frequency component. The noise removal processing circuit according to claim 9, wherein the noise removal processing circuit is less than the degree of parallelism. The wavelet transform circuit that transmits a low-frequency component to another wave red transform circuit includes a two-dimensional two-dimensional wavelet transform circuit as compared with the two-dimensional wavelet transform circuit included in the destination wavelet transform circuit. Or the noise removal processing circuit of Claim 10. When the number of times of performing the wavelet transform is N, the first wavelet transform circuit generates 2 ^(N−1) first low-frequency components from a plurality of lines of pixels, and each wavelet transform circuit The noise removal processing circuit according to claim 7, wherein the wavelet transform is performed at substantially the same timing. Accumulate multiple lines of pixels received in line units,
Generates a plurality of first high-frequency components and a plurality of first low-frequency components by reading the accumulated pixels of the plurality of lines and performing wavelet transform on the respective lines included in the plurality of lines at the same timing. And
A wavelet transform is performed on each of the plurality of first low frequency components at the same timing to generate a second high frequency component and a second low frequency component,
Degenerate a noise component from the plurality of first high frequency components;
Degenerate a noise component from the second high frequency component;
By performing an inverse wavelet transform based on the second high-frequency component and the second low-frequency component in which the noise component is degenerated, a first inverse-transform signal is generated and output,
A second inverse transform signal is generated and output by performing an inverse wavelet transform based on the plurality of first high frequency components in which noise components are degenerated and the first inverse transform signal. A noise removal processing method. A process performed between the first low-frequency component generation process and the second low-frequency component generation process, wherein a plurality of high-frequency components and a plurality of low-frequency components are converted from a plurality of low-frequency components. One or more intermediate wavelet transform processes to generate;
One or more intermediate wavelet reduction processes for reducing noise components from the plurality of high frequency components;
Performing an inverse wavelet transform on the basis of a plurality of high-frequency components with reduced noise components and an inverse transform signal, and performing one or more intermediate inverse wavelet transform processes for generating and outputting a new inverse transform signal;
In the intermediate wavelet transform processing, a plurality of high frequency components and a plurality of low frequency components are generated based on the first low frequency component, and a plurality of high frequency components and a plurality of low frequencies are generated based on the generated plurality of low frequency components. Perform the process of generating the component one or more times,
The intermediate inverse wavelet transform process generates a new inverse transform signal based on the first inverse transform signal and performs a process of generating a new inverse transform signal based on the generated inverse transform signal at least once. The noise removal processing method described.

Images