JP4255071B2 - Pixel value selection type encoding device and decoding device of interest - Google Patents

Description

  The present invention belongs to the field of digital image processing, and particularly relates to a target pixel value selection type encoding apparatus and decoding apparatus suitable for realizing efficient encoding of CT images.

  In CT image data, since the pixel value represents the absorption coefficient of radiation, the pixel value is determined to some extent depending on the part or organ. When performing image diagnosis using CT images, the display brightness from the pixel value is displayed in the viewer so that only the part with a specific CT value is displayed with luminance gradation in order to read the part or organ that the doctor is interested in Display with gradation conversion to value. Since the luminance gradation of a general display system is an 8-bit accuracy per component, the 16-bit or CT-value CT is converted into a 8-bit gradation and displayed.

In general, display is performed by linearly associating the pixel value p (x, y) in the pixel data with the display luminance value I (x, y) by the following equation (1). Only when a pixel value of a certain pixel enters a window (hereinafter referred to as a “display window”), it is displayed with an 8-bit luminance gradation.
I (x, y) = 0: p (x, y) ≦ w _{min
= 255 (p (x, y} ) -w min) / (w max -w min)
: W _min ≦ p (x, y) <w _max
= 255: w _max <p (x, y)
... Formula (1)

Generally, the display window is represented by the window width W (> 0) and the window center L, and the upper and lower limits of the window are calculated as w _min = L−W / 2 and w _max = L + W / 2, respectively. FIG. 13 shows the relationship between the pixel value determined by equation (1) and the display luminance value. In addition, since the organ to be noted is determined to some extent in the captured image, it is possible to specify the window of the pixel value to be noted to some extent.

  Patent Document 1 discloses that a signal is decomposed into conversion coefficients, mask data for distinguishing between conversion data to be transmitted and conversion data not to be transmitted is generated, and the conversion coefficients are hierarchically ordered. Is disclosed.

Further, Patent Document 2 discloses an image processing apparatus that encodes an image of a region including ROI with higher image quality than surrounding images.
JP-T-2001-513608 JP 2002-135782 A

  On the other hand, CT has a large number of images that are generated at one time, and in consideration of accumulation and transmission, more efficient encoding is desired than the current encoding method. There is a concept called ROI (Region of Interest) that preferentially encodes an important part in image diagnosis such as a specific part in an image. The problem with this ROI is how to identify it.

  For this, a method that a doctor designates himself or a method using an automatic image recognition technique is conceivable. However, as long as a spatial region of an image is selected, there is a problem of accuracy of the region specifying method. That is, the ROI includes a portion that is not necessary for image diagnosis at a certain rate.

  On the other hand, for the purpose of improving coding efficiency, only the display gradation image with the bit accuracy reduced to 8 bits, which was used by doctors for image diagnosis, is encoded and stored, greatly reducing the amount of storage. I can expect to do it. However, with such a method, since the encoded image can be decoded and displayed only at a fixed gradation, it is not possible to meet the need for another doctor to convert the gradation and display it later. Of course, it is impossible to return to an image with a complete bit accuracy (12 or 16 bits).

  The present invention has been made in view of the above-described conventional technology, and an object of the present invention is to select a pixel value of interest that can perform high-precision encoding only for necessary pixels and reduce the accumulation amount. It is to provide an encoding device. Another object is to provide a pixel value selection type encoding device that can perform image encoding without reducing the bit accuracy of a high gradation CT image. Another object of the present invention is to provide a decoding device that can return the data encoded by the encoding device to the original bit accuracy.

In order to achieve the above-described object, the present invention provides a frequency conversion unit that performs frequency conversion on an array of pixels of image data, and whether a pixel value exists within a certain range for each pixel of the image data. A target pixel value determination unit that determines whether or not , and a shape conversion unit that performs frequency conversion similar to the frequency conversion on the mask information that distinguishes between the inside and outside of the range determined by the target pixel value determination unit; A partial quantization unit that performs quantization based on a specified quantization ratio on a selected coefficient among transform coefficients output from the frequency conversion unit based on the frequency-transformed mask information An entropy encoding unit that encodes the quantized transform coefficient obtained from the partial quantization unit, data encoded by the entropy encoding unit, the frequency-converted mask information, and the amount There is a first feature data including a reduction ratio that has and a file generation unit for filing.

Further, based on the mask information subjected to frequency conversion, a partial shift unit that shifts a selected specific coefficient among the quantized transform coefficients output from the quantization unit to the significant bit side, and an encoding A second feature is that a target quantized transform coefficient group is divided into bit planes, and further includes a bit plane encoding unit that further divides the bit planes and performs hierarchical encoding.

Further, the partial shift unit sets the shift amount of the selected coefficient to a value greater than or equal to the dynamic range of the original coefficient, and the data generated by the file generation unit does not include the mask information. The point has the third feature.

Further, in the decoding apparatus, a bit plane decoding unit that performs bit plane decoding of the encoded bit string, a shape conversion mask reproduction unit that reproduces a shape conversion mask from the quantized transform coefficient reproduced by the bit plane decoding unit, and the shape A fourth feature is that an inverse partial shift unit that performs inverse shift of the identified coefficient using the shape conversion mask reproduced by the transformation mask reproduction unit is provided.

According to the first to third aspects of the invention, for example, it is possible to perform highly accurate encoding only on a pixel of a portion (for example, a bone or an organ) necessary for image diagnosis or the like. In other words, for example, pixels with a portion that is not necessary for image diagnosis (for example, peripheral portions such as bones and organs) are not completely accurately encoded. For this reason, highly efficient encoding is implement | achieved. In addition, the highly accurate encoding can be performed while maintaining the pixels inside the LOI of the high gradation CT image with high quality.

  In addition, by performing high-precision encoding for pixels and high-compression encoding for other pixels, the compressed data size is reduced without degrading the image quality of the display window of interest, and the overall coding efficiency Can be improved.

  In addition, it is possible to preferentially encode only a necessary target pixel such as an important part in the image diagnosis with high precision, and an unnecessary part is not subjected to high precision encoding at all.

According to the inventions of claims 4 and 5 , the shape conversion mask can be reproduced by itself.

  The present invention will be described below with reference to the drawings. The present invention preferentially encodes a pixel having a specific pixel value or luminance with high accuracy instead of preferentially encoding a pixel existing in a specific region as in the conventional ROI. It proposes a completely new concept of LOI (LOI; Level of Interest). The LOI means a specific “luminance range”.

  FIG. 1 is a block diagram illustrating a configuration of a main part of a target pixel value region selection type encoding device (hereinafter referred to as a LOI encoder) according to a first embodiment of the present invention.

  A width W and a central pixel value L are input to a target pixel value determination unit (hereinafter referred to as a LOI determination unit) 1 to form a LOI window. Also, a quantization ratio is set in the partial quantization unit 2 and the file generation unit 4.

When CT image data a is input, the image data a is input to the LOI determination unit 1 and the partial quantization unit 2. The LOI determination unit 1 determines a target pixel included in the LOI window from the image data a. Here, LOI mask information M (x, y) representing the position (x, y) of a pixel included in the LOI window is created by the following equation.
M (x, y) = 1: w _min ≦ p (x, y) ≦ w _max
= 0: Other than the above The mask information M (x, y) is sent to the partial quantization unit.

The partial quantization unit 2, the quantization precision, performs quantization have coarse in the other pixels with respect to the mask information M (x, y) by determining pixel to be included in the LOI window . For this reason, the L OI does not include any part that is not necessary for image diagnosis.

  The degree of quantization can be input as a quantization ratio from the outside. For example, when the quantization ratio is (2/4), the quantization step width is reduced for the pixel q determined to be included in the LOI window, and conversely for the pixel r determined not to be included. For example, the quantization step width is increased (for example, doubled). As a result, for example, the pixel q can be quantized with higher accuracy than the quantization performed on the pixel r. Therefore, the pixel q included in the LOI window is quantized while maintaining the high quality of the high gradation CT image. Note that by setting the LOI mask to three or more values instead of binary values as in the above equation (see FIG. 2), it is possible to set multistage quantization accuracy.

  The quantized transform coefficient b obtained by the quantization is encoded by the entropy encoding unit 3. The encoded bit string c output from the entropy encoding unit 3 is sent to the file generation unit 4. The file generation unit 4 converts the encoded bit string c into a file together with mask information and a quantization ratio, and outputs the file as an encoded file d.

  By storing this encoded file, the amount of accumulation can be greatly reduced. Further, when decoding is performed by a later-described decoder (FIG. 3), the image of the region of interest (LOI) can be completely restored to the original bit precision image. The same applies to the LOI encoder of each embodiment described later.

  Next, the configuration of a decoder (hereinafter referred to as LOI decoder) for decoding the encoded file d created by the LOI encoder will be described with reference to FIG.

  The file analysis unit 5 analyzes the encoded file d and extracts mask information, a quantization ratio, and an encoded bit string c. The encoded bit string c is decoded by the entropy decoding unit 6 to obtain a quantized transform coefficient b. The partial inverse quantization unit 7 determines whether the pixel is included in the LOI window or not based on the mask information. In the case of the pixel in the LOI window, the pixel in the LOI window is appropriately quantized in each case. Inverse quantization using steps.

  As described above, the pixel e included in the LOI window is reproduced as a quantized image e with high accuracy in pixels inside the LOI of the high gradation CT image and with low accuracy for pixels not included.

  Next, a second embodiment of the present invention will be described with reference to FIG. 4, FIG. 5, and FIG. This embodiment is characterized in that the LOI encoder is applied to encoding using frequency transform such as wavelet transform. In these drawings, the same or equivalent parts as those in FIGS. 1 and 3 are denoted by the same reference numerals.

  The LOI determination unit 1 creates the same LOI mask information as described in the first embodiment. The image data a is frequency-converted by the frequency converter 11 before being encoded, and becomes a conversion coefficient f. The shape conversion unit 12 performs the same frequency conversion on the LOI mask information as described above, and indicates which conversion coefficient f is associated with a pixel included in the LOI window. Create a shape conversion mask. FIG. 5 shows how the shape conversion mask is created. Assuming that the mask information created by the LOI determination unit 1 is given the reference numeral 13 (for example, the portion of k painted in black is 1 and the others are 0), the shape conversion unit 12 The shape conversion mask that has undergone shape conversion is like that denoted by reference numeral 14.

  The partial quantization unit 2 performs the same partial quantization process as described in the first embodiment using this shape conversion mask. Since the processes of the partial quantization unit 2, the entropy encoding unit 3 and the file generation unit 4 are the same as those in the first embodiment, description thereof will be omitted. For this reason, the pixels included in the LOI window are encoded while maintaining the high quality of the high gradation CT image.

  FIG. 6 shows a configuration of the LOI decoder for decoding the encoded file generated by the LOI encoder of FIG. The operations of the file analysis unit 5, the entropy decoding unit 6 and the partial inverse quantization unit 7 of the LOI decoder are the same as or equivalent to the operations of FIG. In the present embodiment, the partial dequantization unit 7 partially dequantizes the quantized transform coefficient b based on the mask information after shape conversion and the quantization ratio, and reproduces the transform coefficient f. The inverse conversion unit 8 reproduces the image e by inversely converting the conversion coefficient f.

  Also in this embodiment, as in the first embodiment, the pixel e included in the LOI window (for example, the pixel included in k in FIG. 5) is reproduced with high accuracy, and the pixel e included in the LOI window is reproduced with low accuracy. .

  Next, a third embodiment of the present invention will be described with reference to FIG. 7, FIG. 8, and FIG. This embodiment is characterized in that a bit plane encoder 23 is used as an entropy encoder and partial quantization accuracy control is realized by a bit shift of a coefficient. In these drawings, the same or equivalent parts as those in FIGS. 4 and 6 are denoted by the same reference numerals.

  The LOI determination unit 1 receives the width W and the center pixel value L, and the partial shift unit 22 and the file generation unit 4 are set with a shift amount from the outside. In addition, the frequency converter 11, the LOI determination unit 1, and the shape converter 12 perform the same operations as described in FIG.

  The transform coefficient f obtained by the frequency transform unit 11 is quantized by the quantizer 21 by a normal method. As a result, the quantized transform coefficient b is obtained, and only the quantized transform coefficient specified using the shape transform mask is shifted, for example, shifted up to the significant bit side by the partial shift unit 22. As a result, the shifted-up quantized coefficient is preferentially encoded in the bit plane encoding unit 23 at the subsequent stage. The state of this bit shift is shown in FIG. As shown in FIG. 8, the quantized transform coefficient 15 (portion filled with diagonal lines) included in the shape transform mask is shifted up as indicated by reference numeral 16.

  Next, the bit plane encoding unit 23 performs encoding from the upper bits to the lower bits. When there is a rate specification 24 from the outside, the generation of the code is terminated when the generated code reaches the specified bit amount. For this reason, high-bit encoding can be performed for important images, and generation of codes can be terminated for non-important images, so that efficient encoding can be performed. For this reason, the pixels included in the LOI are encoded with higher accuracy than the pixels not included in the LOI. The generated encoded bit string c is filed by the file generation unit 4 together with the shape conversion mask information and the shift amount.

  Next, the configuration and operation of the LOI decoder will be described with reference to FIG. In addition, the same code | symbol is attached | subjected to the same or equivalent as FIG.

  The file analysis unit 5 analyzes the encoded file d, and extracts shape-converted mask information, shift amount, and encoded bit string c. The encoded bit string c is decoded by the bit plane decoding unit 25 to obtain a quantized transform coefficient b. In the bit plane decoding unit 25, by specifying the rate 26 from the outside, it is possible to stop decoding when the bit string is decoded by the specified bit amount.

  The inverse partial shift unit 27 determines whether the coefficient is included in the LOI window or not based on the mask information after shape conversion. If the coefficient is in the LOI window, the coefficient is shifted down by the shift amount. That is, in FIG. 8, the quantized transform coefficient 16 is shifted down to the code 15. Thereafter, the inverse quantization unit 28 performs inverse quantization, and the inverse transform unit 8 performs inverse transform to reproduce the image e.

  Also in this embodiment, as in the first and second embodiments, the pixels included in the LOI window are shifted up and encoded and decoded, for example, with 16 bits, so that they are reproduced with high accuracy and are not included. Pixels are reproduced with low accuracy.

  In addition, this embodiment is a hierarchical code that generates a bitstream that is preferentially reproduced with high quality from the pixels included in the LOI at the stage of sequentially decoding the encoded bitstream from the front to the rear. It has a function to convert. Further, since the decoder of FIG. 9 has a bit plane decoding unit, it is possible to reproduce only the pixels included in the LOI by skipping the decoding in the middle.

  Next, a fourth embodiment of the present invention will be described with reference to FIG. 10, FIG. 11, and FIG. In the LOI encoder of this embodiment, the shift amount in the partial shift unit 22 is set to be equal to or greater than the dynamic range of the quantized coefficient so that the information of the shape conversion mask can be decoded without being sent to the decoder side. There are features.

  That is, as shown in FIG. 11, the partial shift unit 22 is specified by using a specific coefficient, for example, a shape conversion mask, by the maximum partial shift method in which the shift amount is greater than or equal to the dynamic range of the quantization coefficient. The quantized transform coefficient 31 (the shaded area) is shifted up to a code 32 that is greater than or equal to the dynamic range. The file generation unit 4 files the encoded bit string c and the shift amount together, but the shape conversion mask information is not included in the encoded file. Since the points other than the above are the same as the operation in FIG.

  Next, the configuration and operation of the LOI decoder for decoding the encoded file generated by the LOI encoder 4 of FIG. 10 will be described with reference to FIG.

  The file analysis unit 5 analyzes the encoded file and extracts the shift amount and the encoded bit string c. The encoded bit string c is decoded by the bit plane decoding unit 25 to obtain a quantized transform coefficient b. In the bit plane decoding unit 25, as in the case of FIG. 9, by specifying the rate 26 from the outside, it is possible to stop decoding when the bit string is decoded by the specified bit amount.

  The shape conversion mask reproducing unit 41 reproduces a shape conversion mask from the quantized coefficients subjected to LOI decoding. In this reproduction, assuming that the number of bits of the dynamic range of the original quantized coefficient that has not been shifted up is B, the portion where the coefficient with significant B bits is significant among the decoded quantized coefficients is the shape conversion LOI mask. Can be done. The inverse partial shift unit 27 shifts down the coefficient specified by the shape conversion mask. That is, the reverse shift of FIG. 11 is performed. The subsequent procedure is the same as that in FIG.

  Also in the present embodiment, as in the third embodiment, at the stage of sequentially decoding the encoded bitstream from the front to the rear, the high-quality reproduction is preferentially performed from the pixels included in the LOI. A hierarchical coding function for generating a simple bitstream. In addition, since the decoder of FIG. 12 has a bit plane decoding unit, it is possible to reproduce only the pixels included in the LOI by stopping decoding halfway.

  Furthermore, since it is not necessary to send shape conversion mask information to the decoder side, the amount of data sent from the encoder to the decoder can be reduced.

It is a block diagram which shows the structure of the LOI encoding apparatus of 1st Embodiment of this invention. It is explanatory drawing of an example of the mask information which a LOI determination part produces. It is a block diagram which shows the structure of the decoding apparatus corresponding to 1st Embodiment. It is a block diagram which shows the structure of the LOI encoding apparatus of 2nd Embodiment of this invention. It is explanatory drawing of an example of the shape conversion mask information which a shape conversion part produces. It is a block diagram which shows the structure of the decoding apparatus corresponding to 2nd Embodiment. It is a block diagram which shows the structure of the LOI encoding apparatus of 3rd Embodiment of this invention. It is explanatory drawing of operation | movement of a partial shift part. It is a block diagram which shows the structure of the decoding apparatus corresponding to 3rd Embodiment. It is a block diagram which shows the structure of the LOI encoding apparatus of 4th Embodiment of this invention. It is explanatory drawing of operation | movement of a partial shift part. It is a block diagram which shows the structure of the decoding apparatus corresponding to 4th Embodiment. It is explanatory drawing of a display window.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... LOI determination part, 2 ... Partial quantization part, 3 ... Entropy encoding part, 4 ... File generation part, 11 ... Frequency conversion part, 12 ... Shape conversion part, 21 Quantization unit,
22... Partial shift unit, 23... Bit plane encoding unit, 25... Bit plane decoding unit, 27... Reverse partial shift unit, 41.

Claims (5)

A frequency conversion unit that performs frequency conversion on an array of pixels of image data;
A target pixel value determination unit that determines whether a pixel value is within a certain range for each pixel of the image data;
A shape conversion unit that performs frequency conversion similar to the frequency conversion on the mask information that distinguishes between the inside and outside of the range determined by the target pixel value determination unit;
A partial quantization unit that performs quantization based on a specified quantization ratio on a selected coefficient among the transform coefficients output from the frequency conversion unit based on the frequency-transformed mask information; ,
An entropy encoding unit that encodes the quantized transform coefficient obtained from the partial quantization unit;
A pixel value selection type encoding comprising: a file generation unit configured to file data including the data encoded by the entropy encoding unit, the frequency-converted mask information, and the quantization ratio apparatus. A frequency conversion unit that performs frequency conversion on an array of pixels of image data;
A quantization unit for quantizing the transform coefficient output from the frequency transform unit;
A target pixel value determination unit that determines whether a pixel value is within a certain range for each pixel of the image data;
A shape conversion unit that performs frequency conversion similar to the frequency conversion on the mask information that distinguishes between the outside and the range determined by the target pixel value determination unit;
Based on the frequency-transformed mask information, a partial shift unit that shifts a selected specific coefficient of quantized transform coefficients output from the quantizing unit to the significant bit side,
A bit plane encoding unit that divides a quantized transform coefficient group to be encoded into bit planes, and further divides the bit planes to perform hierarchical encoding;
A file generation unit configured to file data including data encoded by the bit plane encoding unit, the frequency-converted mask information, and a shift amount;
The target pixel value selection type encoding device, wherein the bit plane encoding unit generates a code having a size substantially equal to a code amount designated from outside at the time of bit plane encoding. The partial shift unit sets the shift amount of the selected coefficient to a value that is greater than or equal to the dynamic range of the original coefficient,
The pixel value selection type encoding device according to claim 2, wherein the data generated by the file generation unit does not include the mask information. A decoding device that decodes encoded data filed from the target pixel value selection type encoding device according to claim 3,
Analyzing the filed encoded data and obtaining a coded bit string and shift amount; and
A bit plane decoding unit that performs bit plane decoding of the encoded bit string;
A shape conversion mask reproducing unit for reproducing a shape conversion mask from the quantized transform coefficient reproduced by the bit plane decoding unit;
Using the shape conversion mask reproduced by the shape conversion mask reproducing unit, an inverse partial shift unit for performing an inverse shift of the identified coefficient;
An inverse quantization unit that inversely quantizes the quantized transform coefficient obtained by the inverse partial shift unit;
A decoding apparatus comprising: an inverse transform unit that inversely transforms the transform coefficient obtained by the inverse quantization unit and reproduces an image.   5. The decoding apparatus according to claim 4, wherein the bit plane decoding unit decodes a code having a size substantially equal to a code amount designated from the outside at the time of bit plane decoding.

Images