JP2007243632A - Signal processing apparatus and method, recording medium, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve a prediction accuracy with respect to an image frequently viewed by a user. <P>SOLUTION: A television receiver 10 obtains feature quantity information of an image corresponding to a received image signal and stores the information to a history information memory 13. A prediction coefficient generating apparatus 20 acquires feature quantity information of an image viewed by the user from the history information memory 13 of a board 12 removed from the television receiver 10 and uses the feature quantity information of the image viewed by the user to produce coefficient seed data used by the television receiver 10. This invention can be applied to image signal processing systems each comprising a television receiver for using a prediction coefficient to convert an SD signal into an HD signal and a prediction coefficient generating apparatus for generating the prediction coefficient. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a signal processing apparatus and method, a recording medium, and a program, and in particular, by generating a prediction coefficient corresponding to an image often viewed by the user, the prediction accuracy for the image often viewed by the user is improved. The present invention relates to a signal processing apparatus and method, a recording medium, and a program that can be made to operate.

  Conventionally, as an image data converter (upconverter) that converts low-resolution image data into high-resolution image data by subjecting low-resolution image data to pixel interpolation by performing frequency interpolation processing using an interpolation filter. After classifying the low-resolution image data into classes according to the signal level distribution of each pixel, the prediction coefficient corresponding to the class is read from a memory in which data called a prediction coefficient is stored in advance. There is an image data conversion apparatus that employs so-called class classification adaptive processing for predicting and calculating high resolution image data from resolution image data.

  Prediction coefficients stored in the memory of the image data converter are generated in advance by data processing called learning. The learning circuit that generates the prediction coefficient uses high-resolution image data as a teacher image using a digital filter that is selected from a plurality of digital filters having different passbands and that is optimal for the high-resolution image data as a teacher image. Is converted into low-resolution image data as a student image, and learning is performed between the high-resolution image data and the low-resolution image data to generate a prediction coefficient ( Patent Document 1).

  For example, in recent years, the development of television receivers that can obtain higher-resolution images has been desired due to the increase in audio / visual orientation. It has been developed. The number of high-definition scan lines is 1125, which is more than twice that of 525 scan lines in the NTSC (National Television Standards Committee) system. Also, the aspect ratio of the high vision is 9:16, whereas the aspect ratio of the NTSC system is 3: 4. For this reason, high-definition images can be displayed with higher resolution and presence than in the NTSC system.

  The television receiver incorporating the above-described image data converter extracts pixel data of a block (area) corresponding to pixel data at a target position of a high-definition video signal from the NTSC video signal, and Based on the level distribution pattern of the pixel data, the class to which the pixel data at the above-mentioned target position belongs is determined, the prediction coefficient is read out corresponding to this class, and the pixel data at the target position is generated using the prediction coefficient. .

  Here, the prediction coefficient of the image data conversion apparatus of Patent Document 1 is obtained by learning by the least square method using an image set of dozens of types of teacher images. At this time, during normal development, an image set with various characteristics is prepared in advance and used as a learning set in order to prevent any input signal from failing. ing.

  That is, it is a feature of the conventional prediction coefficient that exhibits statistically robust performance.

JP 2002-218414 A

  However, the characteristics of the television signal that the user actually watches at home do not always match the characteristics of the image set used when learning the prediction coefficient.

  For example, when the viewing content is extremely biased or the reception state is extremely uncommon, the image data conversion apparatus may not be able to exhibit its original performance with a statistically robust prediction coefficient.

  The present invention has been made in view of such circumstances, and is intended to reduce prediction errors for images often viewed by a user and improve prediction accuracy.

  The signal processing device according to the first aspect of the present invention is a signal processing device that generates a coefficient used for an operation for converting a first image signal into a second image signal having a resolution higher than that of the first image signal. The feature amount storage means for storing the feature amount information of the image corresponding to the first image signal generated by another signal processing device that converts the first image signal into the second image signal From the acquisition unit for acquiring the feature amount information of the image, and the calculation for converting into the second image signal by the other signal processing device using the feature amount information of the image acquired by the acquisition unit. Coefficient generating means for generating coefficients to be used.

  The feature amount storage means may be provided on a substrate that can be attached to and detached from the other signal processing device.

  The feature amount information of the image may be a feature amount frequency distribution of the image.

  Using feature amount generation means for generating feature amount information of a learning image, feature amount information of the learning image generated by the feature amount generation means, and feature amount information of the image acquired by the acquisition means, Image selection means for selecting an image similar to the image from the learning image, and the coefficient generation means uses the learning image selected by the image selection means, by the other signal processing device. It is possible to generate a coefficient used for the calculation for converting to the second image signal.

  A signal processing method according to a first aspect of the present invention is a signal processing device that generates a coefficient used for an operation for converting a first image signal into a second image signal having a higher resolution than the first image signal. In the signal processing method, image feature amount information corresponding to the first image signal generated by another signal processing device that converts the first image signal into the second image signal is stored. Coefficients used for the calculation of acquiring the feature amount information of the image from the feature amount storage means and using the acquired feature amount information of the image to convert to the second image signal by the other signal processing device The step of generating is included.

  The program according to the first aspect of the present invention performs a process of generating a coefficient used for an operation for converting a first image signal into a second image signal having a higher resolution than the first image signal. A feature amount for storing feature amount information of an image corresponding to the first image signal generated by a signal processing device that converts the first image signal into the second image signal Step of acquiring feature amount information of the image from a storage unit, and using the acquired feature amount information of the image, generating a coefficient used for an operation of converting to the second image signal by the signal processing device including.

  The program recorded on the recording medium according to the first aspect of the present invention includes a coefficient used for an operation for converting the first image signal into a second image signal having a higher resolution than the first image signal. A program that causes a computer to perform processing to be generated, the image feature amount corresponding to the first image signal generated by a signal processing device that converts the first image signal into the second image signal The feature amount information of the image is acquired from the feature amount storage means in which the information is stored, and the signal processing device converts the acquired feature amount information of the image into the second image signal using the acquired feature amount information of the image. Generating a coefficient to be used.

  The signal processing device according to the second aspect of the present invention is a signal processing device that converts a first image signal into a second image signal having a resolution higher than that of the first image signal. Feature quantity information generating means for generating feature quantity information of an image corresponding to the first image signal to be converted into a feature quantity storage for storing the feature quantity information of the image generated by the feature quantity information generating means And calculating the first image signal by using the coefficient generated by the other signal processing apparatus using the feature amount information of the image stored in the feature amount storage unit. Image signal converting means for converting to a second image signal.

  The feature amount storage means may be provided on a detachable substrate.

  The feature amount information of the image may be a feature amount frequency distribution of the image.

  The signal processing method according to the second aspect of the present invention is the signal processing method of the signal processing device for converting the first image signal into the second image signal having a higher resolution than the first image signal. Image feature amount information corresponding to the first image signal to be converted into the second image signal is generated, the generated feature amount information of the image is stored in a feature amount storage unit, and the feature amount storage unit A step of converting the first image signal into the second image signal by performing an operation using a coefficient generated by another signal processing apparatus using the stored feature amount information of the image. including.

  A program according to a second aspect of the present invention is a program for causing a computer to perform a process of converting a first image signal into a second image signal having a higher resolution than the first image signal. Image feature amount information corresponding to the first image signal to be converted into the second image signal is generated, the generated feature amount information of the image is stored in a feature amount storage unit, and the feature amount storage unit Including a step of converting the first image signal into the second image signal by performing an operation using a coefficient generated by the signal processing device using the stored feature amount information of the image. .

  The program recorded on the recording medium according to the second aspect of the present invention causes the computer to perform a process of converting the first image signal into a second image signal having a higher resolution than the first image signal. A program that generates feature amount information of an image corresponding to the first image signal to be converted into the second image signal, and stores the generated feature amount information of the image in a feature amount storage unit The first image signal is converted into the second image by performing calculation using the coefficient generated by the signal processing device using the feature amount information of the image stored in the feature amount storage means. Converting to a signal.

  In the first aspect of the present invention, the first image signal generated by the signal processing device that converts the first image signal into a second image signal having a higher resolution than the first image signal is included in the first image signal. The feature amount information of the image is acquired from the feature amount storage means in which the feature amount information of the corresponding image is stored. And the coefficient used for the calculation converted into a said 2nd image signal by the said signal processing apparatus is produced | generated using the acquired feature-value information of the said image.

  In the second aspect of the present invention, image feature amount information corresponding to the first image signal converted into the second image signal is generated, and the generated feature amount information is the feature amount storage means. Is remembered. Then, the first image signal is obtained by performing the calculation using the coefficient generated by the signal processing device using the feature amount information of the image stored in the feature amount storage means. Converted to image signal.

  ADVANTAGE OF THE INVENTION According to this invention, the prediction precision with respect to the image which a user often views can be improved. As a result, the user can view an image with good image quality.

  Embodiments of the present invention will be described below. Correspondences between constituent elements of the present invention and the embodiments described in the specification or the drawings are exemplified as follows. This description is intended to confirm that the embodiments supporting the present invention are described in the specification or the drawings. Therefore, even if there is an embodiment that is described in the specification or the drawings but is not described here as an embodiment corresponding to the constituent elements of the present invention, that is not the case. It does not mean that the form does not correspond to the constituent requirements. Conversely, even if an embodiment is described here as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

  The signal processing device according to the first aspect of the present invention is a signal processing device that generates a coefficient used for an operation for converting a first image signal into a second image signal having a resolution higher than that of the first image signal. (For example, in the prediction coefficient generation device 20 in FIG. 4) the first image signal is generated by another signal processing device (for example, the television reception device 10 in FIG. 4) that converts the first image signal into the second image signal. Acquisition means (for example, the feature amount information of the image) from the feature amount storage means (for example, the history information memory 13 of FIG. 4) in which the feature amount information of the image corresponding to the first image signal is stored. The frequency distribution comparison unit 221 in FIG. 17 and the coefficient used for the calculation to convert the second image signal by the other signal processing apparatus using the feature amount information of the image acquired by the acquisition unit. Coefficient to generate Comprising forming means (e.g., the coefficient seed data production unit 206 of FIG. 17) and.

  The feature amount storage means may be provided on a substrate (for example, the substrate 12 in FIG. 4) that can be attached to and detached from the other signal processing device.

  Feature amount generation means (for example, a learning image frequency distribution generation unit 204 in FIG. 17) that generates feature amount information of a learning image, feature amount information of the learning image generated by the feature amount generation means, and the acquisition means Image selection means (for example, the learning image selection unit 222 in FIG. 17) that selects the learning image similar to the image using the feature amount information of the image acquired by The generation unit can generate a coefficient used for an operation of converting the learning image selected by the image selection unit into the second image signal by the other signal processing device.

  A signal processing method or program according to the first aspect of the present invention is a signal for generating a coefficient used for an operation for converting a first image signal into a second image signal having a higher resolution than the first image signal. In a signal processing method or program of a processing device (for example, the prediction coefficient generation device 20 in FIG. 4), another signal processing device (for example, in FIG. 4) that converts the first image signal into the second image signal. The feature amount of the image is stored in the feature amount storage means (for example, the history information memory 13 in FIG. 4) in which the feature amount information of the image corresponding to the first image signal generated by the television receiver 10) is stored. Information is acquired (for example, step S101 in FIG. 28), and the coefficient used for the calculation for converting to the second image signal by the other signal processing device using the acquired feature amount information of the image. To formed (e.g., step S67 of FIG. 26) includes the step.

  The signal processing device according to the second aspect of the present invention is a signal processing device that converts a first image signal into a second image signal having a resolution higher than that of the first image signal (for example, the television shown in FIG. 4). In the receiving device 10), feature amount information generating means (for example, a feature amount frequency distribution generating unit in FIG. 7) that generates feature amount information of an image corresponding to the first image signal converted into the second image signal. 66), feature amount storage means (for example, history information memory 13 in FIG. 4) for storing the feature amount information of the image generated by the feature amount information generation means, and the feature amount storage means. The first image signal is converted into the first image signal by performing calculation using a coefficient generated by another signal processing apparatus (for example, the prediction coefficient generation apparatus 20 in FIG. 4) using the image feature amount information. Image signal converted to 2 image signal Switching means (for example, estimating the prediction computation unit 58 in FIG. 7) and a.

  The feature amount storage means may be provided on a detachable substrate (for example, the substrate 12 in FIG. 4).

  The signal processing method or program according to the second aspect of the present invention is a signal processing device (for example, as shown in FIG. 4) that converts a first image signal into a second image signal having a higher resolution than the first image signal. In the signal processing method or program of the television receiver 10), image feature amount information corresponding to the first image signal to be converted into the second image signal is generated (for example, step S44 in FIG. 25). The feature amount information of the generated image is stored in the feature amount storage means (for example, the history information memory 13 in FIG. 4) (for example, step S45 in FIG. 25), and the image stored in the feature amount storage means is stored. The feature value information of the first image signal is calculated using the coefficient generated by another signal processing device (for example, the prediction coefficient generation device 20 in FIG. 4), so that the first image signal is converted into the second image signal. Image To convert (e.g., step S20 in FIG. 24) includes the step.

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

  First, the outline of the present invention will be described with reference to FIGS. FIG. 1 shows the concept of learning a conventional prediction coefficient.

  As a conventional learning method of a prediction coefficient used when predicting and calculating high resolution image data from low resolution image data, as shown in FIG. 1, an HD (High Definition) signal consisting of only one scene is used. Dozens of samples from 1 to n are prepared for combinations of images (hereinafter also referred to as HD images) and SD (Standard Definition) signal images (hereinafter also referred to as SD images) generated from the HD images. Prediction coefficients were obtained by the least square method with all of them added.

  That is, the prediction coefficient obtained by the conventional learning method can be expressed by a regression line of Y = aX + b for all samples, where the HD image is the teacher data Y and the SD image is the student data X. . For this reason, the prediction error between the prediction value Y ′ obtained using each SD image and the prediction coefficient Y = aX + b and the prediction coefficient may be large depending on the sample, as indicated by a point on the graph. There was a case.

  FIG. 2 shows the concept of individual learning of prediction coefficients. In the example of FIG. 2, the prediction coefficient is obtained by learning individually for each image as compared to the example of FIG.

  For example, a prediction coefficient obtained by learning with a combination of the HD image 1 and the SD image 1 generated from the HD image 1 can be represented by a regression line of Y = a1X + b1. The prediction error between the prediction value Y ′ obtained using the SD image 1 and the prediction coefficient Y = a1X + b1 and the prediction coefficient is smaller than in the example of FIG.

  Similarly, the prediction coefficient obtained by learning from the combination of the HD image 2 and the SD image 2 generated from the HD image 2 can be represented by a regression line of Y = a2X + b2, and is predicted with the SD image 2 The prediction error between the prediction value Y ′ obtained using the coefficient Y = a2X + b2 and the prediction coefficient is also smaller than in the example of FIG.

  The prediction coefficient obtained by learning from the combination of the HD image n and the SD image n generated from the HD image n can be represented by a regression line of Y = anX + bn. The prediction error between the prediction value Y ′ obtained using Y = anX + bn and the prediction coefficient is also smaller than in the example of FIG.

  That is, as in the example of FIG. 2, the prediction error can be reduced and the prediction accuracy can be improved by obtaining the prediction coefficient according to the characteristics of each image individually. Therefore, if a prediction coefficient can be learned and generated according to an image that the user views at home (hereinafter also referred to as a user viewing image), an apparatus that generates high-resolution data using the prediction coefficient The best performance can be demonstrated.

  However, in reality, it is difficult to obtain all HD signals corresponding to the SD signal viewed by the user. Therefore, as shown in FIG. 3, the feature amount information of the user viewing image and the SD image at the time of prediction coefficient learning is used.

  FIG. 3 shows the concept of the prediction coefficient of the present invention, which is an extension of the method of FIG.

  First, before learning the prediction coefficient, the feature amount frequency distribution 1 is obtained as the feature amount information of the SD image 1 from the SD image 1 generated from the HD image 1, and from the SD image 2 generated from the HD image 2. The feature amount frequency distribution 2 is obtained as the feature amount information of the SD image 2, and the feature amount frequency distribution n is obtained as the feature amount information of the SD image n from the SD image n generated from the HD image n.

  Then, as indicated by the arrow A1, the feature amount frequency distributions 1 to n are compared with the feature amount frequency distribution obtained from the user viewing image, and have features similar to the feature amount frequency distribution of the user viewing image. Only the SD image of the feature amount frequency distribution is selected, and the prediction coefficients are learned and obtained only by the combination of the selected SD image and HD image, as indicated by arrows A2-1 to A2-3. It is done.

  For example, when the feature amount frequency distribution 1 of the SD image 1 has a feature similar to the feature amount frequency distribution of the user viewing image, the SD image 1 is selected and selected as indicated by an arrow A2-1. The prediction coefficient represented by the regression line of Y = a1X + b1 is learned and obtained only by the combination of the SD image 1 and the HD image 1. When the feature amount frequency distribution 2 of the SD image 2 has a feature similar to the feature amount frequency distribution of the user viewing image, the SD image 2 is selected and the selected SD as shown by the arrow A2-2. The prediction coefficient represented by the regression line of Y = a2X + b2 is learned and obtained only by the combination of the image 2 and the HD image 2.

  Similarly, when the feature amount frequency distribution n of the SD image n has a feature similar to the feature amount frequency distribution of the user viewing image, the SD image n is selected and selected as indicated by an arrow A2-3. The prediction coefficient represented by the regression line of Y = anX + bn is learned and obtained only by the combination of the SD image n and the HD image n.

  As described above, even when prediction coefficient learning using the user viewing image itself is difficult, by using only an image having characteristics similar to the user viewing image for learning, a prediction coefficient specialized for the user viewing image is used. Can be generated.

  Therefore, in the television receiver 10 that generates an HD image from an SD image, which will be described below, an image viewed by the user is converted using a prediction coefficient generated specifically for the image. Thereby, the prediction error at the time of generating the HD image is reduced, and the prediction accuracy is improved. As a result, an image viewed by the user can be displayed with an optimum image quality.

  FIG. 4 shows a configuration of an embodiment of an image signal processing system to which the present invention is applied.

  The image signal processing system shown in FIG. 4 is installed in a user's house, receives an image signal supplied from an external broadcasting station, displays it on a built-in display, and performs product maintenance and the like. It is installed on the manufacturer side or the like, and is configured by a prediction coefficient generation device 20 that generates a prediction coefficient for the television receiver 10.

  The television receiver 10 includes an antenna 11 and receives an image signal, which is an SD (Standard Definition) signal such as an NTSC signal, supplied from the outside, via the antenna 11. The television receiver 10 performs image signal processing using class classification adaptive processing on the received image signal, and converts the image signal into an HD (High Definition) signal. Then, the television receiver 10 displays an image corresponding to the converted image signal on the display unit 38 (FIG. 5). Detailed description regarding the image signal processing will be described later.

  In addition, the television receiver 10 obtains the feature amount information of the image corresponding to the image signal received by the antenna 11 that the user desires to view, that is, the image viewed by the user, and stores it in the history information memory 13. ing. The history information memory 13 is configured on a substrate 12 that is detachably attached to the housing of the television receiver 10. The board 12 is removed from the television receiver 10 and connected to the prediction coefficient generator 20 when generating a prediction coefficient for the television receiver 10.

  The prediction coefficient generation device 20 acquires feature amount information of the user viewing image that is an image viewed by the user from the history information memory 13 of the board 12 removed from the television receiving device 10, and the feature amount information of the user viewing image Is used to generate coefficient seed data used by the television receiver 10. The generated coefficient seed data is provided to the television receiver 10 via the board 12, another board, or a network (not shown).

  That is, in the conversion to the HD signal of the television receiver 10, the coefficient seed data generated by the prediction coefficient generator 20 using the feature amount information of the user viewing image obtained by itself is used. The description of the coefficient seed data will be described later.

  FIG. 5 is a block diagram illustrating a detailed configuration example of the television receiver 10 of FIG.

  The television receiver 10 obtains an SD signal called a 525i signal from a broadcast signal, converts the 525i signal into an HD signal called a 1050i signal, and displays an image based on the HD signal.

  FIG. 6 shows the pixel position relationship of a frame (F) with a 525i signal and a 1050i signal. The pixel position of the odd (o) field is indicated by a solid line, and the pixel position of the even (e) field is indicated by a broken line. ing. Large dots are pixels of 525i signal, and small dots are pixels of 1050i signal. As can be seen from FIG. 6, the pixel data of the 1050i signal includes line data L1, L1 ′ at positions close to the line of the 525i signal and line data L2, L2 ′ at positions far from the line of the 525i signal. Here, L1 and L2 are line data of odd fields, and L1 'and L2' are line data of even fields. The number of pixels in each line of the 1050i signal is twice the number of pixels in each line of the 525i signal.

  Returning to FIG. 5, the configuration of the television receiver 10 will be described. The user operates the television receiver 10 using the remote controller 41. The television receiver 10 includes a microcontroller including a CPU (Central Processing Unit), a RAM (Random Access Memory), and a ROM (Read Only Memory), a system controller 32 for controlling the operation of the entire system, and a remote The communication unit 31 communicates with the controller 41. The communication unit 31 is connected to the system controller 32, receives a remote control signal output from the remote controller 41 according to a user operation, and supplies an operation signal corresponding to the signal to the system controller 32. Has been.

  The antenna 11 receives a broadcast signal (RF (Radio Frequency) modulation signal). The tuner 33 receives a broadcast signal received via the antenna 11 and selects a channel selected by the user using the remote controller 41 in accordance with a control signal input from the system controller 32. In addition, the above-described SD signal (525i signal) is obtained by performing intermediate frequency amplification processing, detection processing, and the like. The buffer memory 34 temporarily stores the SD signal output from the tuner 33.

  The image signal processing unit 35 performs image signal processing for converting an SD signal (525i signal) temporarily stored in the buffer memory 34 into an HD signal (1050i signal). The history information memory 13 described above is configured in the image signal processing unit 35.

  FIG. 7 is a block diagram showing a more detailed configuration of the image signal processing unit 35.

  The first tap selection unit 51, the second tap selection unit 52, and the third tap selection unit 53 of the image signal processing unit 35 use the HD signal from the SD signal (525i signal) stored in the buffer memory 34. Data of a plurality of SD pixels located around the target position in the signal (1050i signal) is selectively extracted and output.

  The first tap selection unit 51 selectively extracts data of SD pixels used for prediction (hereinafter also referred to as “prediction tap”). The second tap selection unit 52 selectively extracts data of SD pixels (hereinafter also referred to as “space class taps”) used for class classification corresponding to the level distribution pattern of the SD pixel data. The third tap selection unit 53 selectively extracts data of SD pixels (hereinafter also referred to as “motion class taps”) used for class classification corresponding to motion. When the space class is determined using SD pixel data belonging to a plurality of fields, motion information is also included in this space class.

  The space class detection unit 54 detects the level distribution pattern of the space class tap data (SD pixel data) selectively extracted by the second tap selection unit 52, and detects the space class based on the level distribution pattern. The class information of the space class is output.

  In the space class detection unit 54, for example, an operation is performed to compress each SD pixel data from 8-bit data to 2-bit data. The space class detection unit 54 outputs compressed data corresponding to each SD pixel data as class information of the space class. In the present embodiment, data compression is performed by ADRC (Adaptive Dynamic Range Coding). In addition to ADRC, DPCM (predictive coding), VQ (vector quantization), or the like may be used as the information compression means.

  Originally, ADRC is an adaptive requantization method developed for high-performance coding for VTR (Video Tape Recorder), but it can express local patterns of signal level efficiently with a short word length. It is suitable for use in data compression. When using ADRC, the maximum value of space class tap data (SD pixel data) is MAX, the minimum value is MIN, the dynamic range of space class tap data is DR (= MAX-MIN + 1), and the number of requantization bits If P is P, the requantized code qi as compressed data is obtained by the calculation of Expression (1) for each SD pixel data ki as the space class tap data. However, in the formula (1), [] means a truncation process. When there are Na SD pixel data as the space class tap data, i = 1 to Na.

  The motion class detection unit 55 mainly detects a motion class for representing the degree of motion from the motion class tap data (SD pixel data) selectively extracted by the third tap selection unit 53. Output class information of motion class.

  In the motion class detection unit 55, the inter-frame difference is calculated from the motion class tap data (SD pixel data) mi and ni selectively extracted by the third tap selection unit 53, and the absolute value of the difference is further calculated. A threshold value process is performed on the average value of the motion vectors, and a motion class that is an index of motion is detected. In other words, the motion class detection unit 55 calculates the average value AV of the absolute value of the difference using the equation (2). In the third tap selection unit 53, for example, as described above, when 12 pieces of SD pixel data m1 to m6 and n1 to n6 are extracted, Nb in Expression (2) is 6.

  Then, in the motion class detection unit 55, the average value AV calculated as described above is compared with one or a plurality of threshold values to obtain class information MV of the motion class. For example, when three threshold values th1, th2, and th3 (th1 <th2 <th3) are prepared and four motion classes are detected, MV = 0 and th1 <AV ≦ th2 when AV ≦ th1. , MV = 1, th2 <AV ≦ th3, MV = 2, and th3 <AV, MV = 3.

  The class synthesis unit 56 should be created based on the re-quantization code qi as the class information of the space class output from the space class detection unit 54 and the class information MV of the motion class output from the motion class detection unit 55. A class code CL indicating the class to which the pixel data (pixel data of the target position) of the HD signal (1050i signal) belongs is obtained.

  In the class synthesis unit 56, the class code CL is calculated by the following equation (3). In Equation (3), Na represents the number of space class tap data (SD pixel data), and P represents the number of requantization bits in ADRC.

  The coefficient memory 57 stores, for each class, a plurality of coefficient data Wi used in the estimation formula used in the estimated prediction calculation unit 58 described later. The coefficient data Wi is information for converting an SD signal (525i signal) into an HD signal (1050i signal). The class code CL output from the class synthesis unit 56 is supplied to the coefficient memory 57 as read address information. The coefficient memory 57 receives coefficient data Wi (i = 1 to n) of the estimation formula corresponding to the class code CL. ) Is read out and supplied to the estimated prediction calculation unit 58.

  Further, the image signal processing unit 35 has an information memory bank 61. In the estimated prediction calculation unit 58 to be described later, the HD pixel data to be created from the prediction tap data (SD pixel data) xi and the coefficient data Wi read from the coefficient memory 57 by the estimation expression of the following expression (4). y is calculated. In Expression (4), n represents the number of prediction taps selected by the first tap selection unit 51.

  Here, the positions of the n pixel data of the prediction tap selectively extracted by the first tap selection unit 51 are in the spatial direction (horizontal and vertical directions) and the temporal direction with respect to the target position in the HD signal. It is over.

  Then, the coefficient data Wi (i = 1 to n) of the estimation formula is generated by a generation formula including parameters r and z as shown in the following formula (5). The information memory bank 61 stores coefficient seed data w10 to wn9, which are coefficient data in this generation formula, for each class. A method for generating the coefficient seed data will be described later.

  As described above, when the 525i signal is converted into the 1050i signal, it is necessary to obtain four pixels of the 1050i signal corresponding to one pixel of the 525i signal in each of the odd and even fields. In this case, the four pixels in the 2 &times; 2 unit pixel block constituting the 1050i signal in each of the odd and even fields have different phase shifts with respect to the central prediction tap.

  FIG. 8 shows a phase shift from the center prediction tap SD0 in the four pixels HD1 to HD4 in the 2 &times; 2 unit pixel block constituting the 1050i signal in the odd field. Here, the positions of HD1 to HD4 are shifted from the position of SD0 by k1 to k4 in the horizontal direction and m1 to m4 in the vertical direction, respectively.

  FIG. 9 shows a phase shift from the center prediction tap SD0 ′ in the four pixels HD1 ′ to HD4 ′ in the 2 &times; 2 unit pixel block constituting the 1050i signal in the even field. Here, the positions of HD1 'to HD4' are shifted from the position of SD0 'by k1' to k4 'in the horizontal direction and m1' to m4 'in the vertical direction, respectively.

  Accordingly, the information memory bank 61 stores coefficient seed data w10 to wn9 for each combination of class and output pixels (HD1 to HD4, HD1 'to HD4').

  The coefficient generation unit 62 uses the coefficient seed data of each class and the values of the parameters r and z, and calculates the estimation expression corresponding to the values of the parameters r and z for each class according to the above-described equation (5). Coefficient data Wi (i = 1 to n) is generated. The coefficient generation unit 62 is loaded with the above-described coefficient seed data of each class from the information memory bank 61. Further, the values of parameters r and z are supplied from the system controller 32 to the coefficient generator 62.

  The coefficient data Wi (i = 1 to n) of each class generated by the coefficient generation unit 62 is stored in the coefficient memory 57 described above. The generation of coefficient data Wi of each class in the coefficient generation unit 62 is performed, for example, in each vertical blanking period. Thereby, even if the values of the parameters r and z are changed by the user's operation of the remote controller 41, the coefficient data Wi of each class stored in the coefficient memory 57 corresponds to the values of the parameters r and z. It can be changed immediately, and noise and resolution can be adjusted smoothly by the user.

  The normalization coefficient calculation unit 63 calculates the normalization coefficient S corresponding to the coefficient data Wi (i = 1 to n) obtained by the coefficient generation unit 62 using the following equation (6). The normalization coefficient memory 64 stores this normalization coefficient S. The normalization coefficient memory 64 is supplied with the class code CL output from the class synthesis unit 56 described above as read address information, and the normalization coefficient S corresponding to the class code CL is read from the normalization coefficient memory 64. And supplied to a normalization calculation unit 59 described later.

  The estimated prediction calculation unit 58 uses the prediction tap data (SD pixel data) xi selectively extracted by the first tap selection unit 51 and the coefficient data Wi read out from the coefficient memory 57 to obtain the equation (4). The pixel data of the HD signal to be created (pixel data at the target position) is calculated using the estimation formula.

  As described above, when an SD signal (525i signal) is converted to an HD signal (1050i signal), four pixels of the HD signal (HD1 to HD4 in FIG. 8, HD1 in FIG. 9) are converted to one pixel of the SD signal. Therefore, the estimated prediction calculation unit 58 generates pixel data for each 2 &2; unit pixel block constituting the HD signal. That is, in the estimated prediction calculation unit 58, the first tap selection unit 51 configures prediction tap data xi corresponding to four pixels (target pixel) in the unit pixel block, and the coefficient memory 57 forms the unit pixel block. The coefficient data Wi corresponding to the four pixels to be supplied is supplied, and the data y1 to y4 of the four pixels constituting the unit pixel block are individually calculated by the estimation formula of the above-described formula (4).

  The normalization calculation unit 59 reads out the four-pixel data y1 to y4 sequentially output from the estimated prediction calculation unit 58 from the normalization coefficient memory 64, and uses coefficient data Wi (i = 1 to Normalize by dividing by the normalization coefficient S corresponding to n). As described above, the coefficient generation unit 62 obtains the coefficient data Wi of the estimation formula, but the obtained coefficient data includes a rounding error, and the sum of the coefficient data Wi (i = 1 to n) becomes 1.0. It is not guaranteed to be. For this reason, the data y1 to y4 of each pixel calculated by the estimated prediction calculation unit 58 has a level fluctuated due to a rounding error. Therefore, normalization by the normalization calculation unit 59 can remove the level fluctuation.

  The post-processing unit 60 line-sequentially converts the four-pixel data y1 'to y4' in the unit pixel block, which are normalized and sequentially supplied by the normalization calculation unit 59, and outputs them in the format of 1050i signal.

  When a user operation signal is input from the communication unit 31, the system controller 32 supplies the notification to the information generation control unit 65.

  The information generation control unit 65 controls the processing of the feature amount frequency distribution generation unit 66 at the timing when the input of the user operation signal is notified from the system controller 32. That is, the information generation control unit 65 controls the feature amount frequency distribution generation unit 66, and whenever the user operates the remote controller 41 such as channel selection, volume adjustment, resolution adjustment, or input switching, for example, The feature amount information of the image of the SD signal temporarily stored in the buffer memory 34 is obtained.

  The feature amount frequency distribution generation unit 66 receives the SD signal (525i signal) stored in the buffer memory 34, and generates the feature amount information of the image of the input SD signal for each combination of the motion class and the space class. The feature amount information of the image for each combination of the generated motion class and space class is recorded in the history information memory 13. For example, an image feature amount frequency distribution is obtained as image feature amount information.

  In the history information memory 13, feature frequency distribution data of a plurality of images generated each time a user operation signal is input, that is, every time the user operates the remote controller 41, includes a motion class and a space class. It is memorized every time. In the example of FIG. 7, the illustration is omitted, but in the television receiver 10, the history information memory 13 is configured on a removable substrate 12.

  Not only the history information memory 13 but also the information generation control unit 65 and the feature amount frequency distribution generation unit 66 may be added to be configured on the substrate 12. Further, not only the history information memory 13 but also the information memory bank 61 may be added to be configured on the substrate 12, and the entire image signal processing unit 35 is detachable in order to make it possible to upgrade the function. It can also be constructed on a flexible substrate 12.

  FIG. 10 is a block diagram illustrating a detailed configuration example of the feature amount frequency distribution generation unit 66.

  The motion class tap selection unit 71, the space class tap selection unit 72, and the feature amount generation tap selection unit 73 of the feature amount frequency distribution generation unit 66 are stored in the buffer memory 34 under the control of the information generation control unit 65. From the SD signal, data of a plurality of SD pixels located around the target position in the HD signal (1050i signal) is selectively extracted and output.

  The motion class tap selection unit 71 is configured basically in the same manner as the third tap selection unit 53 of FIG. 7, and selectively selects SD pixels used for class classification corresponding to motion, that is, motion class tap data. To take out. The space class tap selection unit 72 is basically configured in the same manner as the second tap selection unit 52 of FIG. 7, and is used for class classification corresponding to a level distribution pattern of SD pixel data, that is, a space class tap. The data is selectively extracted. When the space class is determined using SD pixel data belonging to a plurality of fields, motion information is also included in this space class.

  The feature amount generation tap selection unit 73 selectively extracts data of SD pixels (hereinafter also referred to as “feature amount generation taps”) used for generation of feature amounts. The feature value generation tap may have the same tap structure as the space tap or a different tap structure.

  The motion class detection unit 74 is configured in the same manner as the motion class detection unit 55 of FIG. 7, and is mainly based on motion class tap data (SD pixel data) selectively extracted by the motion class tap selection unit 71. A motion class for representing the degree of the motion class is detected, and class information (MV) of the motion class is output.

  The space class detection unit 75 is configured in the same manner as the space class detection unit 54 of FIG. 7, and detects the level distribution pattern of the space class tap data (SD pixel data) selectively extracted by the space class tap selection unit 72. Then, based on the level distribution pattern, a space class is detected, and class information (requantized code qi) of the space class is output.

  The feature amount generation unit 76 generates a feature amount of the SD pixel from the feature amount generation tap data (SD pixel data) selectively extracted by the feature amount generation tap selection unit 73, and the feature amount value (data ) Is output. Note that the motion class detected by the motion class detector 74 is also detected for each pixel with respect to the SD signal to which the spatial class detected by the space class detector 75 is input. A feature amount is generated for each of the same pixels.

  FIG. 11 shows an example of the pixel values of the pixels xo to x6 in the horizontal direction and feature amounts generated from these pixel values. The maximum value of the pixel values of the pixels xo to x6 in the example of FIG. 11 is Max, and the minimum value is Min.

  For example, it is assumed that seven pixels of pixels xo to x6 are selected as feature amount generation taps and the feature amount of the pixel x3 is obtained. At this time, a value that can be generated using the pixel values of the seven pixels xo to x6 is the feature amount of the pixel x3.

  That is, as shown in FIG. 11, the in-tap maximum value Max of the pixel values of the pixels xo to x6 and the pixel value of the pixels xo to x6 generated using the pixel values of the seven pixels xo to x6. The minimum value Min within the tap, the dynamic range DR of the pixel values of the pixels xo to x6, the intermediate value Mean of the pixel values of the pixels xo to x6, or the average value Ave of the pixel values of the pixels xo to x6 is a pixel x3. It is used as a feature amount.

  In addition, it is possible to generate a plurality of types of the maximum value Max within the tap, the minimum value Min within the tap, the dynamic range DR, the intermediate value Mean, and the average value Ave, and use them as the feature amount of the pixel. Only one type can be generated and used as a feature amount of a pixel.

  The feature amount frequency distribution calculation unit 77 corresponds to the same pixel, the motion class detected by the motion class detection unit 74, the space class detected by the space class detection unit 75, and the feature generated by the feature amount generation unit 76. Based on the quantity, feature quantity information, that is, feature quantity frequency distribution (histogram) is generated. At this time, the feature amount frequency distribution is generated for each combination of the motion class and the space class input simultaneously, and the data is stored in the history information memory 13 for each combination of the motion class and the space class.

  FIG. 12 shows an example of pixel feature value and feature frequency distribution. In the example of FIG. 12, the feature amount values of the pixels a1 to a12 among the plurality of pixels constituting the image and the feature amount frequency distribution generated from these feature amount values are shown. The feature amount frequency distribution is configured with the horizontal axis as the feature amount value and the vertical axis as the frequency.

  For example, since the feature amount value of the pixel a1 and the pixel a12 is 50, the feature amount frequency distribution calculation unit 77 counts the frequency of the feature amount value 50 in the feature amount frequency distribution by two. Since the feature amount values of the pixels a2 to a4 and the pixel a7 are 100, the feature amount frequency distribution calculation unit 77 counts the frequency of the feature amount value 100 in the feature amount frequency distribution by four. Since the feature amount value of the pixel a5 is 240, the feature amount frequency distribution calculating unit 77 counts the frequency of the feature amount value 240 in the feature amount frequency distribution by one.

  Since the feature amount value of the pixel a6 and the pixel a11 is 80, the feature amount frequency distribution calculating unit 77 counts the frequency of the feature amount value 80 in the feature amount frequency distribution by two. Since the feature amount value of the pixel a8 is 20, the feature amount frequency distribution calculating unit 77 counts the frequency of the feature amount value 20 in the feature amount frequency distribution by one. Since the feature amount value of the pixel a9 is 200, the feature amount frequency distribution calculation unit 77 counts the frequency of the feature amount value 200 in the feature amount frequency distribution by one. Since the feature amount value of the pixel a10 is 10, the feature amount frequency distribution calculation unit 77 counts the frequency of the feature amount value 10 in the feature amount frequency distribution by one.

  This feature amount frequency distribution is generated for each combination of motion class and space class input simultaneously, and the data is recorded in the history information memory 13 for each combination of motion class and space class.

  FIG. 13 shows an example of the feature amount frequency distribution of each user viewing image stored in the history information memory 13.

  In the history information memory 13, as shown in FIG. 13, every time a user operation signal from the remote controller 41 is input, an image viewed by the user (that is, a user viewing image) generated from the SD signal is input. Feature amount frequency distribution data is stored for each combination of motion class and space class.

  For example, every time a user operation signal such as an operation signal corresponding to the user's channel selection, an operation signal corresponding to the user's input switching, or an operation signal corresponding to the user's resolution adjustment is input from the remote controller 41. The feature frequency distribution data of the user viewing images 1 to n generated in the above are stored for each combination of motion classes 0 to P and space classes 0 to N, respectively.

  That is, for each of the user viewing images 1 to n, motion class 0 and space class 0 feature quantity frequency distribution data, motion class 0 and space class 1 feature quantity frequency distribution data, motion class 0 and space class 2 Feature quantity frequency distribution data,..., Motion class 0 and space class P feature quantity frequency distribution data, motion class 1 and space class 0 feature quantity frequency distribution data, motion class 2 and space class 0 feature quantities (P + 1) × (N + 1) each, such as frequency distribution data, motion class N and space class 0 feature amount frequency distribution data,..., Motion class P and space class N feature amount frequency distribution data Is stored.

  In the above description, the case where the feature amount information is generated every time the operation signal from the remote controller 41 is input has been described. However, the feature amount information can be generated every predetermined time, or a specific operation signal is input. It can also be generated only when

  Further, in the feature quantity frequency distribution generation unit 66 of FIG. 7, the motion class tap selection unit 71 and the motion class detection unit 74, and the space class tap selection unit 72 and the space class detection unit 75 configured in the example of FIG. The third tap selection unit 53 and the motion class detection unit 55, and the second tap selection unit 52 and the space class detection unit 54 in the image signal processing unit 35 of FIG. You can also.

  Returning to FIG. 5, the configuration of the television receiver 10 will be described again.

  An OSD (On Screen Display) processing unit 36 generates a display signal SCH for displaying a character graphic on the screen of the display unit 38. The combining unit 37 combines the display signal SCH output from the OSD processing unit 36 with the HD signal output from the image signal processing unit 35 and supplies the combined signal to the display unit 38. The display unit 38 is configured by a flat panel display such as a CRT (cathode-ray tube) display or LCD (liquid crystal display), for example, and an image based on the HD signal output from the image signal processing unit 35 and, if necessary. The display signal SCH combined by the combining unit 37 is displayed.

  In addition, a drive 39 is connected to the system controller 32 as necessary, and a removable medium 40 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is appropriately mounted, and a computer program read from them is loaded. Installed in the system controller 32 as necessary.

  The operation of the television receiver 10 in FIG. 5 will be described.

  The system controller 32 controls the tuner 33 based on a user operation input using the remote controller 41. The tuner 33 performs channel selection processing, intermediate frequency amplification processing, detection processing, and the like on the broadcast signal received by the antenna 11 under the control of the system controller 32, and outputs the result to the buffer memory 34.

  The SD signal (525i signal) output from the tuner 33 is supplied to the buffer memory 34 and temporarily stored. The SD signal temporarily stored in the buffer memory 34 is supplied to the image signal processing unit 35 and converted into an HD signal (1050i signal) based on a control signal supplied from the system controller 32.

  That is, the image signal processing unit 35 obtains pixel data constituting the HD signal (hereinafter referred to as “HD pixel data”) from pixel data constituting the SD signal (hereinafter referred to as “SD pixel data”). Can do. The HD signal output from the image signal processing unit 35 is combined with a character graphic or the like by the display signal SCH output from the OSD processing unit 36 in the combining unit 37 as necessary, and is supplied to the display unit 38. An image is displayed on the screen of the display unit 38.

  Further, the user can adjust the horizontal and vertical resolutions of the image displayed on the screen of the display unit 38 and the noise removal degree (noise reduction degree) by operating the remote controller 41. In the image signal processing unit 35, HD pixel data is calculated by the estimation formula. As coefficient data of the estimation formula, parameters r and horizontal resolution that are adjusted by the user's operation of the remote controller 41 and r are determined. The one corresponding to the parameter z that determines the degree of noise removal is generated and used by a generation formula including these parameters r and z. As a result, the horizontal and vertical resolution and noise removal degree of the image by the HD signal output from the image signal processing unit 35 correspond to the adjusted parameters r and z.

  FIG. 14 shows an example of a user interface for adjusting the parameters r and z. At the time of adjustment, an adjustment screen 81 in which the adjustment positions of the parameters r and z are indicated by an asterisk icon 82 in the drawing is displayed on the display unit 38 as an OSD. The remote controller 41 includes a joystick 91 as user operation means.

  The user can move the position of the icon 82 on the adjustment screen 81 by operating the joystick 91, and adjust the values of the parameter r that determines the horizontal and vertical resolutions and the parameter z that determines the degree of noise removal. can do.

  FIG. 15 shows an enlarged view of the adjustment screen 81 shown in FIG. The value of parameter z that determines the degree of noise removal is adjusted by moving icon 82 left and right, while the value of parameter r that determines the horizontal and vertical resolution is adjusted by moving icon 82 up and down. The user can easily adjust the values of the parameters r and z with reference to the adjustment screen 81 displayed on the display unit 38.

  Note that the remote controller 41 may include other pointing devices such as a mouse and a trackball instead of the joystick 91. Further, the values of the parameters r and z adjusted by the user may be displayed numerically on the adjustment screen 81.

  Next, the operation of the image signal processing unit 35 described above will be described with reference to FIG.

  The second tap selection unit 52 is supplied with the SD signal (525i signal) stored in the buffer memory 34 and receives four pixels (attention position) in the unit pixel block constituting the HD signal (1050i signal) to be created. The space class tap data (SD pixel data) located around the (pixel) is selectively extracted. The space class tap data (SD pixel data) selectively extracted by the second tap selection unit 52 is supplied to the space class detection unit 54. The space class detection unit 54 performs ADRC processing on each SD pixel data as space class tap data, and re-creates the space class (mainly class classification for waveform representation in space) as class information. A quantization code qi is obtained (see equation (1)).

  The third tap selection unit 53 is supplied with the SD signal (525i signal) stored in the buffer memory 34 and receives four pixels (unit pixels) constituting the HD signal (1050i signal) to be created. Data of motion class taps (SD pixel data) located around the pixel at the target position) is selectively extracted. The motion class tap data (SD pixel data) selectively extracted by the third tap selection unit 53 is supplied to the motion class detection unit 55. The motion class detection unit 55 obtains class information MV of a motion class (mainly class classification for representing the degree of motion) from each SD pixel data as motion class tap data.

  The motion information MV and the requantization code qi are supplied to the class synthesis unit 56. For each unit pixel block constituting the HD signal (1050i signal) to be created from the supplied motion information MV and the requantization code qi, the class synthesis unit 56 includes four pixels (target pixel) in the unit pixel block. ) To obtain the class code CL indicating the class to which it belongs (see equation (3)). The class code CL is supplied to the coefficient memory 57 and the normalized coefficient memory 64 as read address information.

  For example, in each vertical blanking period, for each combination of class and output pixels (HD1 to HD4, HD1 ′ to HD4 ′) corresponding to the values of parameters r and z adjusted by the user in the coefficient generation unit 62. The coefficient data Wi (i = 1 to n) of the estimation formula is obtained using the coefficient seed data w10 to wn9 and stored in the coefficient memory 57 (see formula (5)). Also, the normalization coefficient S corresponding to the coefficient data Wi (i = 1 to n) of the estimation formula obtained by the coefficient generation unit 62 is generated by the normalization coefficient calculation unit 63 and stored in the normalization coefficient memory 64. (See equation (6)).

  By supplying the class code CL to the coefficient memory 57 as read address information, four output pixels corresponding to the class code CL from the coefficient memory 57 (HD1 to HD4 in the odd field, HD1 ′ to HD4 ′ in the even field). The coefficient data Wi of the estimation equation for the minute is read and supplied to the estimated prediction calculation unit 58. Further, the first tap selection unit 51 receives the SD signal (525i signal) stored in the buffer memory 34, and receives four pixels (unit pixels) constituting the HD signal (1050i signal) to be generated ( Predictive tap data (SD pixel data) located around the pixel of interest position) is selectively extracted.

  The estimated prediction calculation unit 58 uses the prediction tap data (SD pixel data) xi and the coefficient data Wi for four output pixels read from the coefficient memory 57, and the 4 in the unit pixel block constituting the HD signal to be created. Data y1 to y4 of the pixels (pixels at the target position) are calculated (see Expression (4)). The four-pixel data y1 to y4 in the unit pixel block constituting the HD signal sequentially output from the estimated prediction calculation unit 58 is supplied to the normalization calculation unit 59.

  As described above, the class code CL is supplied to the normalization coefficient memory 64 as read address information. From the normalization coefficient memory 64, the normalization coefficient S corresponding to the class code CL, that is, from the estimated prediction calculation unit 58, is supplied. The normalization coefficient S corresponding to the coefficient data Wi used for the calculation of the output HD pixel data y1 to y4 is read and supplied to the normalization calculation unit 59. The normalization calculation unit 59 normalizes the HD pixel data y1 to y4 output from the estimated prediction calculation unit 58 by dividing by the corresponding normalization coefficient S. Thereby, the level fluctuation of the data y1 to y4 due to the rounding error when the coefficient generation unit 62 obtains the coefficient data Wi is removed.

  As described above, the data y1 ′ to y4 ′ of the four pixels in the unit pixel block which are normalized by the normalization calculation unit 59 and sequentially output are supplied to the post-processing unit 50. The post-processing unit 50 line-sequentially converts the four-pixel data y1 'to y4' in the unit pixel block sequentially supplied from the normalization calculation unit 59 and outputs the data in the format of 1050i signal. That is, the post-processing unit 50 outputs a 1050i signal as an HD signal.

  As described above, the image signal processing unit 35 calculates the HD pixel data y using the coefficient data Wi (i = 1 to n) of the estimation formula corresponding to the adjusted parameters r and z. It is. Therefore, the user can freely adjust the horizontal and vertical resolutions of the image by the HD signal and the noise removal degree by adjusting the values of the parameters r and z. In addition, since the coefficient data of each class corresponding to the adjusted values of the parameters r and z is generated and used by the coefficient generation unit 62 each time, a large amount of coefficient data is stored. Memory is not necessary, and memory saving can be achieved.

  Next, the operation of the feature amount frequency distribution generation unit 66 in FIG. 10 will be described.

  For example, when an operation signal corresponding to user tuning, an operation signal corresponding to user input switching, or an operation signal corresponding to user resolution adjustment is input from the remote controller 41, the system controller 32 A notification that each operation signal is input is supplied to the information generation control unit 65.

  When an operation signal input notification is supplied from the system controller 32, the information generation control unit 65 starts processing to the motion class tap selection unit 71, the space class tap selection unit 72, and the feature quantity generation tap selection unit 73. Let

  The motion class tap selection unit 71 is supplied with the SD signal stored in the buffer memory 34, and the four pixels in the unit pixel block constituting the HD signal (1050i signal) to be generated from the SD signal (525i signal). Data of motion class taps (a plurality of SD pixels) located around (a pixel at the target position) is selectively extracted. The motion class tap data selectively extracted by the motion class tap selection unit 71 is supplied to the motion class detection unit 74.

  The motion class detection unit 74 uses the SD pixel data as the motion class tap data to move the SD pixel motion class (mainly, four pixels in the unit pixel block constituting the HD signal (1050i signal) to be created. Class classification for representing the degree of motion) is detected, and class information (MV) of the detected motion class of the SD pixel is output to the feature amount frequency distribution calculation unit 77.

  Also, the space class tap selection unit 72 receives the SD signal (525i signal) stored in the buffer memory 34, and in the unit pixel block constituting the HD signal (1050i signal) to be created from the SD signal. Data of space class taps (a plurality of SD pixels) located around four pixels (the pixel at the target position) is selectively extracted. The space class tap data selectively extracted by the space class tap selection unit 72 is supplied to the space class detection unit 75.

  The space class detection unit 75 performs ADRC processing on each SD pixel data as space class tap data, and is surrounded by four pixels in a unit pixel block constituting an HD signal (1050i signal) to be created. A requantization code (qi) as class information of the SD pixel spatial class (mainly class classification for waveform expression in the space) is obtained and output to the feature quantity frequency distribution calculation unit 77.

  Further, the feature quantity generation tap selection unit 73 is supplied with the SD signal (525i signal) stored in the buffer memory 34, and in the unit pixel block constituting the HD signal (1050i signal) to be created from the SD signal. The data of the feature value generation taps (a plurality of SD pixels) located around the four pixels (the pixel at the target position) are selectively extracted. The feature value generation tap data selectively extracted by the feature value generation tap selection unit 73 is supplied to the feature value generation unit 76.

  The feature quantity generation unit 76 uses the SD pixel data as the feature quantity generation tap data selectively extracted by the feature quantity generation tap selection unit 73 in the unit pixel block constituting the HD signal (1050i signal) to be created. For example, a maximum value and a minimum value, which are feature amounts of the SD pixels surrounded by the four pixels, are generated, and the generated feature value values of the SD pixels are output to the feature amount frequency distribution calculation unit 77.

  Information about the motion class, the space class, and the feature amount obtained for the same SD pixel is supplied to the feature amount frequency distribution calculation unit 77. The feature quantity frequency distribution calculation unit 77 counts the feature quantity frequency for each combination of the motion class and the space class, generates a feature quantity frequency distribution (histogram), and uses the generated feature quantity frequency distribution data as the motion class. And it records in the history information memory 13 for each combination of space classes.

  In this way, the feature frequency distribution data of the user viewing image that is recorded in the history information memory 13 every time an operation signal is input is, for example, in the case of version upgrade of the television receiver 10 or the like. This is used when generating the coefficient seed data w10 to wn9 stored in the information memory bank 61.

  That is, the coefficient seed data w10 to wn9 stored in the information memory bank 61 is generated by the prediction coefficient generation apparatus 20 using the feature amount frequency distribution data of the user viewing image recorded in the history information memory 13. It is.

  Next, an example of a method for generating the coefficient seed data w10 to wn9 will be described. In this example, an example is shown in which coefficient seed data w10 to wn9, which are coefficient data in the generation formula of the above-described formula (5), are obtained.

  Here, for the following description, ti (i = 0 to 9) is defined as in Expression (7).

  When this equation (7) is used, equation (5) can be rewritten as the following equation (8).

  Finally, the undetermined coefficient wij is obtained by learning. That is, for each combination of class and output pixel, a coefficient value that minimizes the square error is determined using a plurality of SD pixel data and HD pixel data. This is a so-called least square method. Assuming that the learning number is m, the residual in the k-th (1 ≦ k ≦ m) learning data is ek, and the sum of squared errors is E, using Eq. (4) and Eq. ). Here, xik represents the kth pixel data at the i-th predicted tap position of the SD image, and yk represents the corresponding pixel data of the kth HD image.

  In the solution by the least square method, wij is obtained such that the partial differentiation by wij in equation (9) becomes zero. This is shown by the following formula (10).

  Hereinafter, when Xipjq and Yip are defined as in Expression (11) and Expression (12), Expression (10) can be rewritten using a matrix as in Expression (13).

  This equation is generally referred to as a normal equation. The normal equation is solved for wij by using a sweeping method (Gauss-Jordan elimination method) or the like, and coefficient seed data is calculated.

  FIG. 16 is a diagram illustrating a concept of an example of the above-described coefficient seed data generation method.

  A plurality of SD signals are generated from the HD signal. For example, by varying the parameter r for varying the vertical and horizontal bands of the filter used when generating the SD signal from the HD signal and the parameter z for varying the noise amount in 9 stages, a total of 81 Types of SD signals are generated. Learning is performed between the plurality of SD signals and HD signals generated in this way, and coefficient seed data is generated.

  FIG. 17 is a block diagram showing a configuration of the prediction coefficient generation device 20 of FIG.

  The prediction coefficient generation device 20 inputs dozens of kinds of HD signals as learning images and image numbers from a data server (not shown) and the like, and the history information memory 13 of the board 12 removed from the television receiver 10. The coefficient seed data w10 to wn9 stored in the information memory bank 61 of the television receiver 10 is generated using the data of the feature amount frequency distribution which is the feature amount information of the user viewing image stored in.

  The prediction coefficient generation device 20 generates the feature amount frequency distribution of a plurality of learning images, and records the learning image frequency distribution generation unit 204 that records the feature amount frequency distribution of the plurality of learning images, and the feature of the user viewing image stored in the history information memory 13. A learning image selection unit 205 that compares the amount frequency distribution with the feature amount frequency distribution of the learning image recorded by the learning image frequency distribution generation unit 204 and selects a learning image that is similar to the user viewing image. A coefficient seed data generation unit 206 that learns a prediction coefficient using a learning image selected by the learning image selection unit 205, and a microcontroller including a CPU, a RAM, and a ROM, for controlling the operation of the entire system The system controller 207 is used.

  The image number of the learning image input from the input terminal 202 is input to the input terminal 201. The input terminal 202 is inputted with HD signals (1050i signals) of several tens of kinds of learning images as teacher signals in order. The values of the parameters r and z are input to the input terminal 203. Note that the learning image, the image number, and the parameters r, z, and the like are input from an external data server (not shown), for example, under the control of the system controller 207.

  The learning image frequency distribution generation unit 204 includes an SD signal generation unit 211, a feature amount frequency distribution generation unit 212, and a learning image frequency distribution memory 213. The HD signal and its image number input from the input terminals 201 and 202, respectively. In response to the image number, motion class, and space class requested by the learning image selection unit 205, the feature amount frequency distribution, which is the feature amount information of the learning image, is generated and stored for each motion class and space class. Then, the process of outputting the stored feature value frequency distribution data is performed.

  The SD signal generation unit 211 performs horizontal and vertical thinning processing on the HD signal (1050i signal) from the input terminal 202 using the values of the parameters r and z input from the input terminal 203, An SD signal (525i signal) is obtained as a student signal.

  At this time, the SD signal generation unit 211 sets the band in the spatial direction (horizontal and vertical) of the band limiting filter used when generating the SD signal from the HD signal based on the parameter r input from the input terminal 203. The noise addition state for the SD signal is changed stepwise based on the parameter z.

  Therefore, in the SD signal generation unit 211, the bandwidth in the spatial direction is varied according to the input parameter r value itself, and the noise addition state is varied stepwise according to the input parameter r value itself. Is done.

  The feature amount frequency distribution generation unit 212 is basically configured in the same manner as the feature amount frequency distribution generation unit 66 of FIG. 7, and calculates the feature amount frequency distribution as the feature amount information of the image of the SD signal from the SD signal generation unit 211. The image is generated for each motion class and space class, and the image feature frequency distribution data for each generated motion class and space class is recorded in the learning image frequency distribution memory 213.

  In the learning image frequency distribution memory 213, the image number from the input terminal 201 and the feature amount frequency distribution data of dozens of types of learning images corresponding to the image number generated by the feature amount frequency distribution generation unit 212 are stored. Stored for each motion class and space class.

  FIG. 18 is a block diagram illustrating a detailed configuration example of the feature amount frequency distribution generation unit 212.

  The motion class tap selection unit 251, the space class tap selection unit 252, and the feature amount generation tap selection unit 253 of the feature amount frequency distribution generation unit 212 are target pixels that generate feature amounts from the SD signal from the SD signal generation unit 211. The data of a plurality of SD pixels located in the periphery of the image are selectively extracted and output. The motion class tap selection unit 251, the space class tap selection unit 252, and the feature amount generation tap selection unit 253 are the motion class tap selection unit 71, the space class tap selection unit 72, and the feature amount generation tap selection unit of FIG. This is basically the same as 73.

  The motion class detection unit 254 is basically configured in the same manner as the motion class detection unit 74 in FIG. 10, and is based on the motion class tap data (SD pixel data) selectively extracted by the motion class tap selection unit 251. Then, a motion class for representing the degree of motion is detected, and class information (MV) of the motion class is output.

  The space class detection unit 255 is configured basically in the same manner as the space class detection unit 75 of FIG. 10, and the level distribution of the space class tap data (SD pixel data) selectively extracted by the space class tap selection unit 252. A pattern is detected, a space class is detected based on the level distribution pattern, and class information (requantized code qi) of the space class is output.

  The feature quantity generation unit 256 is basically configured in the same manner as the feature quantity generation unit 76 of FIG. 10, and is based on the feature quantity generation tap data (SD pixel data) selectively extracted by the feature quantity generation tap selection unit 253. , The feature amount of the SD pixel corresponding to the feature amount generated by the feature amount generation unit 76 of FIG. That is, as described above with reference to FIG. 11, the pixel value of the feature value generation tap is used, and among the maximum value Max within the tap, the minimum value Min within the tap, the dynamic range DR, the intermediate value, and the average value Ave. Is generated as a feature value of the pixel to be obtained.

  The feature amount frequency distribution calculation unit 257 is basically configured in the same manner as the feature amount frequency distribution calculation unit 77 of FIG. 10, and corresponds to the same pixel, and is detected by the motion class detection unit 74 and the space class detection unit. Based on the space class detected from 75 and the feature quantity generated by the feature quantity generator 76, the feature quantity frequency distribution (histogram) described above with reference to FIG. At this time, the feature amount frequency distribution is generated for each combination of the motion class and the space class input at the same time, and the data is stored in the learning image frequency distribution memory 213 for each combination of the motion class and the space class.

  FIG. 19 shows an example of the feature amount frequency distribution stored in the learning image frequency distribution memory 213.

  The learning image frequency distribution memory 213 stores data of feature quantity frequency distributions of several tens of types of learning images for each combination of motion class and space class.

  For example, feature amount frequency distribution data of learning images 1 to m is stored for each combination of motion classes 0 to P and space classes 0 to N, respectively.

  That is, for each learning image 1 to m, feature class frequency distribution data of motion class 0 and space class 0, feature class frequency distribution data of motion class 0 and space class 1, motion class 0 and space class 2 Feature quantity frequency distribution data,..., Motion class 0 and space class P feature quantity frequency distribution data, motion class 1 and space class 0 feature quantity frequency distribution data, motion class 2 and space class 0 feature quantity frequencies Distribution data, motion class N and space class 0 feature quantity frequency distribution data,..., Motion class P and space class N feature quantity frequency distribution data, and so on (P + 1) × (N + 1) Data of feature amount frequency distribution is stored.

  Returning to FIG. 17, the learning image selection unit 205 includes a frequency distribution comparison unit 221, a learning image selection unit 222, and a learning image table 223, and the feature amount frequency distribution of the user viewing image in the history information memory 13 and the learning image. Compare the feature quantity frequency distribution of all learning images in the frequency distribution memory 213, select learning images having similar features for each combination of motion class and space class, record the image number, and generate coefficient data Sometimes, a process of providing the recorded image number is performed.

  The frequency distribution comparison unit 221 uses the feature amount frequency distribution of the user viewing image stored in the history information memory 13 of the board 12 and all the learning in the learning image frequency distribution memory 213 for the same combination of motion class and space class. The image feature amount frequency distribution is compared.

  FIG. 20 is a diagram for explaining a comparison target of the feature amount frequency distribution by the frequency distribution comparison unit 221.

  The history information memory 13 stores feature frequency distribution data of user-viewed images generated by the television receiver 10. In the case of the example of FIG. 20, the data of the feature class frequency distribution of the motion class P and the space class M of the user viewing image 1,..., The data of the feature class frequency distribution of the motion class Q and the space class N of the user viewing image 1. , Data of the feature class frequency distribution of the motion class P and the space class M of the user viewing image 2,..., Data of the feature class frequency distribution of the motion class Q and the space class N of the user viewing image 2,. Data on the feature amount frequency distribution of the viewing image is stored for each combination of the motion class and the space class.

  In addition, the learning image feature amount frequency distribution memory 213 stores data of frequency distributions of dozens of types of learning images generated by the learning image frequency distribution generation unit 204. In the case of the example in FIG. 20, the feature frequency distribution data of the motion class P and the space class M of the learning image 1,..., The feature amount frequency distribution data of the motion class Q and the space class N of the learning image 1,. Feature amount frequency distribution data of motion class P and space class M of image 2,..., Feature amount frequency distribution data of motion class Q and space class N of learning image 2,. Frequency distribution data is stored for each combination of motion class and space class.

  The frequency distribution comparison unit 221 averages the feature amount frequency distribution data of all user viewing images stored in the history information memory 13 for each combination of motion class and space class, and for each combination of motion class and space class. The feature amount frequency distribution of all learning images stored in the learning image feature amount frequency distribution memory 213 is compared for each combination of motion class and space class. To go.

  Here, as shown in FIG. 21, the frequency distribution comparison unit 221 compares the two feature quantity frequency distributions by obtaining a correlation coefficient using values obtained from the two feature quantity frequency distributions.

  FIG. 21 shows a specific example of the comparison method of the feature amount frequency distribution by the frequency distribution comparison unit 221. In the example of FIG. 21, a graph of the feature amount frequency distribution of the learning image and the feature amount frequency distribution of the user viewing image is shown in order from the left. In each graph, the vertical axis represents the learning image frequency fsty (x) and the user viewing image frequency fusr (x), and the horizontal axis represents the feature value (x).

  The frequency distribution comparison unit 221 calculates the frequency average Mfsty and the frequency standard deviation SDfsty from the frequency fsty (x) of the learning image, calculates the frequency average Mfusr and the frequency standard deviation SDfusr from the frequency fusr (x) of the user viewing image, By comparing the correlation coefficient value C (−1 ≦ C ≦ 1) obtained by the following equation (14), the two feature quantity frequency distributions are compared.

  Note that N represents the number of samples (number of pixels for counting the feature frequency distribution). Further, the correlation coefficient value C takes a value of −1 to 1, and the larger the value, the more similar the shape of the feature amount frequency distribution.

  Therefore, in the frequency distribution comparison unit 221, the feature amount frequency distribution of the user viewed images of the motion class P and the space class M after the averaging and the motion class P and the space class M of the learning image 1 shown in FIG. The feature amount frequency distributions are compared, and the comparison coefficient value C, the image number of the learning image 1, the motion class P, and the space class M are output to the learning image selection unit 222. Further, the feature amount frequency distributions of the user viewing images of the motion class P and the space class M after the averaging are compared with the feature amount frequency distributions of the motion class P and the space class M of the learning image 2, and the comparison value is a phase value. The relational numerical value C, the image number of the learning image 2, the motion class P, and the space class M are output to the learning image selection unit 222.

  Although not shown, the learning image 3 and the subsequent images are also compared with the feature amount frequency distribution of the user viewed images of the motion class P and the space class M after averaging, and the correlation coefficient value C, which is the comparison value, is compared. The image numbers after the learning image 3, the motion class P, and the space class M are output to the learning image selection unit 222.

  Similarly, in the frequency distribution comparison unit 221, the feature amount frequency distribution of the user viewed images of the motion class Q and the space class N after the averaging, and the motion class Q and the space class N of the learning image 1 shown in FIG. And the correlation coefficient value C, which is the comparison value, the image number of the learning image 1, the motion class Q, and the space class N are output to the learning image selection unit 222. Further, the feature amount frequency distributions of the user viewed images of the motion class Q and the space class N after the averaging are compared with the feature amount frequency distributions of the motion class Q and the space class N of the learning image 2, and the comparison value is a phase value. The relational numerical value C, the image number of the learning image 2, the motion class Q, and the space class N are output to the learning image selection unit 222.

  Although not shown, the learning image 3 and the subsequent images are also compared with the feature amount frequency distribution of the user viewed images of the motion class Q and the space class N after averaging, and the correlation coefficient value C, which is the comparison value, is compared. The image numbers after the learning image 3, the motion class Q, and the space class N are output to the learning image selection unit 222.

  The learning image selection unit 222 selects one learning image based on the correlation coefficient value C, which is a comparison value of all learning images, for each combination of motion class and space class, and for each combination of motion class and space class. The image number of the selected learning image (hereinafter also referred to as the selected image number) is recorded in the learning image table 223.

  The learning image table 223 stores selected image numbers for each combination of all motion classes and space classes.

  FIG. 22 is a diagram illustrating an example of the learning image selection of the learning image selection unit 222 and the configuration of the learning image table.

  The learning image selection unit 222 receives, from the frequency distribution comparison unit 221, a user of feature frequency distributions for all learning image numbers 1 to L, all motion classes 1 to M, all space classes 1 to N, and combinations thereof. The comparison value of the comparison result with the viewing image is input.

  For example, in the learning image selection unit 222 in FIG. 22, the learning image number “1”, the learning class number “1”, the motion class “1”, the space class “1”, and the comparison value “C_1_1_1” Is input, learning image number “1”, motion class “1”, space class “2”, comparison value “C_1_1_2”,..., Learning image number “1”, motion class “1”, space class “ N ”and the comparison value“ C_1_1_N ”are input. The learning image number “1”, the motion class “2”, the space class “1”, and the comparison value “C_1_2_1” are input, and the learning image number “1”, the motion class “2”, the space class “2”, and the comparison A value “C_1_2_2” is input, and a learning image number “1”, a motion class “2”, a space class “N”, and a comparison value “C_1_2_N” are input.

  Although not shown in the figure, the motion classes “3” to “M−1” and thereafter are input in the same manner, and the learning image number “1”, the motion class “M”, and the space class “1”. The comparison value “C_1_M_1” is input, the learning image number “1”, the motion class “M”, the space class “2”, the comparison value “C_1_M_2” is input,..., The learning image number “1”, the motion class “ M, space class “N”, and comparison value “C_1_M_N” are input, learning image number “1”, motion class “M”, space class “1”, and comparison value “C_1_M_1” are input, and learning image number “ 1 ”, motion class“ M ”, space class“ 2 ”, and comparison value“ C_1_M_2 ”are input,..., Learning image number“ 1 ”, motion class“ M ”, space class“ N ”, and comparison value“ C_1_M_N ” Is entered.

  Thereafter, although illustration thereof is omitted, the learning image numbers “2” to “L”, the learning image numbers “2” to “L”, the motion classes “1” to “M”, and the space class “1” are also omitted. Through “N” and comparison values “C_2_1_1” through “C_L_M_N” are input.

  First, the learning image selection unit 222 rearranges them for each combination of the motion class and the space class, evaluates the comparison value for each combination of the motion class and the space class, and learns the learning image number of the comparison value that is most similar. Is recorded in the learning image table 223 as a selected image number in the combination of the motion class and the space class.

  Note that it is necessary to select a learning image having a feature amount frequency distribution that is most similar to the feature amount frequency distribution of the user viewing image. In this case, the comparison value is described above with reference to FIG. Since the correlation value C is used, the learning image having the largest comparison value is selected as being most similar.

  That is, the comparison values “C_1_1_1” to “C_L_1_1” of the image numbers “1” to “L” in the combination of the motion class “1” and the space class “1” are sequentially arranged, and the comparison values “C_1_1_1” to “C_L_1_1” are arranged. Among them, the image number of the learning image having the largest comparison value is selected and recorded in the learning image table 223 as the selected image number “X_1_1” of the motion class “1” and the space class “1”.

  The comparison values “C_1_1_2” to “C_L_1_2” of the image numbers “1” to “L” in the combination of the motion class “1” and the space class “2” are sequentially arranged, and the comparison values “C_1_1_2” to “C_L_1_2” The image number of the learning image having the maximum comparison value is selected and is recorded in the learning image table 223 as the selected image number “X_1_2” of the motion class “1” and the space class “2”.

  Although not shown, the comparison values of the motion class “1” and the spatial classes “3” to “N−1” are arranged in order in the same manner for the space classes “3” to “N−1” and thereafter. The image number of the learning image having the largest comparison value among the comparison values is selected, and learning is selected as the selection image number of the motion class “1” and the space classes “3” to “N−1”, respectively. It is recorded in the image table 223.

  Similarly, the comparison values “C_1_1_N” to “C_L_1_N” of the image numbers “1” to “L” in the combination of the motion class “1” and the space class “N” are sequentially arranged, and the comparison values “C_1_1_N” to “C_L_1_N” are sequentially arranged. The learning image having the largest comparison value is selected and recorded in the learning image table 223 as the selected image number “X_1_N” of the motion class “1” and the space class “N”.

  Thereafter, although illustration thereof is almost omitted, for the motion classes “2” to “M”, the comparison values “1” to “L” in the combinations with the space classes “1” to “N” “ C_1_2_1 ”to“ C_L_M_N ”are arranged in order, and the image number of the learning image having the largest comparison value among these comparison values is selected, and the motion classes“ 2 ”to“ M ”and the space classes“ 3 ”to“ 3 ”are selected. The selected image numbers “X_2_1” to “X_M_N” of “N−1” are recorded in the learning image table 223, respectively.

  Returning to FIG. 17, the coefficient seed data generation unit 206 includes an image set storage unit 231, an SD signal generation unit 232, a prediction coefficient learning unit 233, and a coefficient seed memory 234, and the motion class and space from the learning image table 223. Using the HD signal of the learning image based on the selected image number for each class, coefficient seed data as a prediction coefficient is generated and recorded in the coefficient seed memory 234.

  The image set storage unit 231 receives and stores several tens of types of learning image HD signals from the input terminal 202. The image set storage unit 231 receives the selected image number from the learning image table 223, selects and outputs the HD signal of the learning image corresponding to the selected image number.

  The SD signal generation unit 232 receives the parameters r and z input from the input terminal 203 for the HD signal (1050i signal) of the learning image selected by the image set storage unit 231 (hereinafter also referred to as a selection learning image). Is used to obtain a SD signal (525i signal) as a student signal by performing horizontal and vertical thinning processing. The configuration of the SD signal generation unit 232 is basically the same as the configuration of the SD signal generation unit 211 described above.

  The prediction coefficient learning unit 233 includes the motion class and space class information from the learning image table 223 corresponding to the selected image number input to the image set storage unit 231, and the HD of the selected learning image from the image set storage unit 231. The signal and the SD signal of the selected learning image from the SD signal generation unit 232 are input.

  For each combination of motion class and space class from the learning image table 223, the prediction coefficient learning unit 233 learns a prediction coefficient using the HD signal and SD signal of the selected learning image corresponding to the combination, and is obtained by learning. The coefficient seed data is recorded in the coefficient seed memory 234 for each combination of motion class and space class.

  FIG. 23 shows a detailed configuration example of the prediction coefficient learning unit 233.

  The first tap selection unit 271, the second tap selection unit 272, and the third tap selection unit 273 of the prediction coefficient learning unit 233 are configured using the SD signal (525i signal) output from the SD signal generation unit 232 as an HD. Data of a plurality of SD pixels located around the target position in the signal (1050i signal) is selectively extracted and output. The first tap selection unit 271 to the third tap selection unit 273 are basically the same as the first tap selection unit 51 to the third tap selection unit 53 of the image signal processing unit 35 described with reference to FIG. It is configured in the same way.

  The space class detection unit 274 detects the level distribution pattern of the space class tap data (SD pixel data) selectively extracted by the second tap selection unit 272, and detects the space class based on the level distribution pattern. And output the class information. The space class detection unit 273 is basically configured in the same manner as the space class detection unit 54 of the image signal processing unit 35 described with reference to FIG. From this space class detection unit 274, a requantization code qi of each SD pixel data as space class tap data is output as class information indicating the space class.

  The motion class detection unit 275 detects a motion class mainly representing the degree of motion from the motion class tap data (SD pixel data) selectively extracted by the third tap selection unit 273, and the class Output information MV. The motion class detection unit 275 is basically configured in the same manner as the motion class detection unit 55 of the image signal processing unit 35 described with reference to FIG. In this motion class detection unit 275, the inter-frame difference is calculated from the motion class tap data (SD pixel data) selectively extracted by the third tap selection unit 273, and the average absolute value of the difference is further calculated. Is subjected to threshold processing, and a motion class as a motion index is detected.

  The first class synthesis unit 276 is based on the re-quantization code qi as the class information of the space class output from the space class detection unit 274 and the class information MV of the motion class output from the motion class detection unit 275. The first class code CL1 indicating the class to which the pixel of interest related to the HD signal (1050i signal) belongs is obtained. The first class synthesis unit 276 is also configured basically in the same manner as the class synthesis unit 56 of the image signal processing unit 35 described with reference to FIG.

  The second class synthesis unit 277 obtains the second class code CL2 from the motion class information MV and the space class information qi input from the learning image table 223.

  The normal equation generation unit 278 includes an image set storage unit according to the first class code CL1 output from the first class synthesis unit 276 and the second class code CL2 output from the second class synthesis unit 277. Each HD pixel data y as pixel data of the target position obtained from the HD signal from 231, and prediction tap data selectively extracted by the first tap selection unit 271 corresponding to each HD pixel data y (SD pixel data) From the values of xi and the parameters r and z, a normal equation (see equation (13)) for obtaining coefficient seed data w10 to wn9 is generated for each class.

  Note that in the normal equation generation unit 278, in order to select one learning image for each class and generate a prediction coefficient, the output of the first class synthesis unit 276, which is the SD signal class detection result, and the learning image A normal equation is generated only when two of the outputs of the second class synthesis unit 277, which are selection results in the table 223, that is, the first class code CL1 and the second class code CL2 match.

  In this case, learning data is generated by a combination of one HD pixel data y and n prediction tap data (SD pixel data) xi corresponding thereto, but the adjusted parameters r and z are adjusted. Corresponding to the change in the value, the spatial band and noise amount in the SD signal generation unit 232 are varied, and a plurality of SD signals are sequentially generated to generate learning data between the HD signal and each SD signal. Is done. As a result, the normal equation generation unit 278 generates a normal equation in which many learning data having different values of the parameters r and z are registered, and the coefficient seed data w10 to wn9 can be obtained.

  Further, in this case, learning data is generated by a combination of one HD pixel data y and n prediction tap data (SD pixel data) xi corresponding thereto, but the normal equation generation unit 278 outputs pixel data. For each (see HD1 to HD4 in FIG. 8, HD1 ′ to HD4 ′ in FIG. 9), a normal equation is generated. For example, a normal equation corresponding to HD1 is generated from learning data composed of HD pixel data y in which the deviation value with respect to the center prediction tap has the same relationship as HD1.

  The coefficient seed data determination unit 279 is supplied with data of normal equations generated for each combination of class and output pixel by the normal equation generation unit 278, solves the normal equation, and for each combination of class and output pixel, The coefficient seed data w10 to wn9 are obtained. The coefficient seed data determination unit 279 obtains coefficient seed data by solving a normal equation by, for example, a sweeping method.

  The coefficient seed memory 234 stores the coefficient seed data obtained by the coefficient seed data determination unit 279. The coefficient seed data is provided to the television receiver 10 via the board 12, another board, or a network (not shown).

  Next, the operation of the learning image frequency distribution generation unit 204 shown in FIG. 18 of the prediction coefficient generation device 20 in FIG. 17 will be described.

  The input terminal 101 is sequentially supplied with HD signals (1050i signals) as teaching signals of several tens of types of learning images, and the SD signal generation unit 211 performs horizontal and vertical thinning processing on the HD signals. Is performed, and SD signals (525i signals) as student signals of tens of kinds of learning images are generated.

  In this case, the SD signal generator 211 generates a parameter r that determines the spatial direction band of the band limiting filter used when generating the SD signal from the HD signal, and a parameter z that determines the noise addition state, in other words, is generated. A parameter r for determining the spatial resolution of the SD signal and a parameter z for determining the amount of noise are input.

  Then, in the SD signal generation unit 211, the spatial direction of the band limiting filter used when generating the SD signal from the HD signal is changed according to the value of the parameter r, and according to the input value of the parameter r itself. The noise addition state is varied step by step.

  The parameters r and z input to the SD signal generator 211 are sequentially changed to change the spatial band and noise addition state of the band limiting filter used when generating the SD signal from the HD signal. As a result, a plurality of SD signals in which the spatial band and the noise addition state change stepwise are generated.

  The motion class tap selection unit 251 of the feature amount frequency distribution generation unit 212 is supplied with the SD signal from the SD signal generation unit 211, and constitutes an HD signal (1050i signal) to be created from the SD signal (525i signal). Data of motion class taps (a plurality of SD pixels) located around the four pixels (pixel at the target position) in the pixel block are selectively extracted. The motion class tap data selectively extracted by the motion class tap selection unit 251 is supplied to the motion class detection unit 254.

  The motion class detection unit 254 uses the SD pixel data as the motion class tap data to move the SD pixel motion class (mainly, four pixels in the unit pixel block constituting the HD signal (1050i signal) to be created. Class classification for expressing the degree of motion) is detected, and class information (MV) of the detected motion class of the SD pixel is output to the feature amount frequency distribution calculation unit 257.

  The space class tap selection unit 252 receives the SD signal (525i signal) from the SD signal generation unit 211, and the four pixels in the unit pixel block constituting the HD signal (1050i signal) to be generated from the SD signal. Data of space class taps (a plurality of SD pixels) located around (a pixel at the target position) is selectively extracted. The space class tap data selectively extracted by the space class tap selection unit 252 is supplied to the space class detection unit 255.

  The space class detection unit 255 performs ADRC processing on each SD pixel data as space class tap data, and is surrounded by four pixels in a unit pixel block constituting an HD signal (1050i signal) to be created. A requantization code (qi) as class information of the SD pixel spatial class (mainly class classification for waveform expression in the space) is obtained and output to the feature quantity frequency distribution calculation unit 257.

  Further, the feature quantity generation tap selection unit 253 receives the SD signal (525i signal) from the SD signal generation unit 211, and the 4 in the unit pixel block constituting the HD signal (1050i signal) to be created from the SD signal. Data of feature quantity generation taps (a plurality of SD pixels) located around the pixel (the pixel at the target position) is selectively extracted. The feature value generation tap data selectively extracted by the feature value generation tap selection unit 253 is supplied to the feature value generation unit 256.

  The feature quantity generation unit 256 uses the SD pixel data as the feature quantity generation tap data selectively extracted by the feature quantity generation tap selection unit 253, in the unit pixel block constituting the HD signal (1050i signal) to be created. For example, a maximum value and a minimum value, which are feature amounts of the SD pixel surrounded by the four pixels, are generated, and the generated feature value of the SD pixel is output to the feature amount frequency distribution calculation unit 257.

  Each information of the motion class, the space class, and the feature amount obtained for the same SD pixel is supplied to the feature amount frequency distribution calculation unit 257. The feature amount frequency distribution calculation unit 257 counts the frequency of feature amounts for each combination of the motion class and the space class from the feature amount frequency distribution calculation unit 257, and generates a feature amount frequency distribution (histogram). Then, the feature amount frequency distribution calculation unit 257 records the generated feature amount frequency distribution data in the learning image frequency distribution memory 213 for each combination of the motion class and the space class in association with the image number from the input terminal 201. To do.

  In this way, the data of the feature amount frequency distribution of several tens of types of learning images is recorded in the learning image frequency distribution memory 213 in advance.

  Next, the operation of the learning image selection unit 205 of the prediction coefficient generation device 20 in FIG. 17 will be described.

  For example, when the board 12 removed from the television receiver 10 is connected to the prediction coefficient generation apparatus 20, the frequency distribution comparison unit 221 of the learning image selection unit 205 is stored in the history information memory 13 of the board 12. The feature frequency distribution data of the user viewing image is read out for each combination of the motion class and the space class, averaged to obtain one feature frequency distribution for each combination of the motion class and the space class, and then the learning image feature The feature frequency distributions of all learning images stored in the quantity frequency distribution memory 213 are compared for each combination of motion class and space class.

  That is, the frequency distribution comparison unit 221 reads the feature amount frequency distribution data of the user viewing image for each combination of the motion class and the space class from the history information memory 13, and averages it for each combination of the motion class and the space class. .

  Then, the frequency distribution comparison unit 221 sequentially reads the feature amount frequency distribution data of all the learning images and the image numbers for each combination of the motion class and the space class, and averages the feature amount frequency distribution of the read learning image. The correlation coefficient value C with the feature amount frequency distribution of the user viewing image thus obtained is compared (see formula (14)).

  The obtained correlation coefficient value C as a comparison value is output to the learning image selection unit 222 together with the corresponding motion class and space class information and the image number.

  The learning image selection unit 222 evaluates the correlation coefficient value C, which is a comparison value of all learning images, for each combination of the motion class and the space class, and the feature amount frequency distribution of the learning image having the maximum correlation coefficient value C is obtained. The learning image is selected as being most similar to the feature amount frequency distribution of the user viewing image, and the selected selected image number is recorded in the learning image table 223 for each combination of the motion class and the space class.

  In this way, in the learning image table 223, the image number of the learning image for which the feature amount frequency distribution is most similar to the user viewing image for each combination of the motion class and the space class is selected image number. Is remembered as

  The selected image number is input to the image set storage unit 231 when the coefficient seed data generation unit 206 generates coefficient seed data of the television receiver 10. At this time, the motion class information and the space class information corresponding to the selected image number are also input to the prediction coefficient learning unit 233.

  Next, the operation of the coefficient seed data generation unit 206 shown in FIG. 23 of the prediction coefficient generation apparatus 20 in FIG. 17 will be described.

  The image set storage unit 231 receives and stores dozens of types of HD signals of learning images that are the same as those input to the learning image frequency distribution generation unit 204 from the input terminal 202. From the learning image table 223, information on the motion class for generating coefficient seed data and information on the space class are sequentially output to the second class synthesis unit 277. At the same time, the selected image number of the motion class and the space class output to the second class synthesis unit 277 is output to the image set storage unit 231.

  The image set storage unit 231 receives the selected image number from the learning image table 223 and selects the HD signal of the learning image corresponding to the selected image number. The HD signal of the learning image selected by the image set storage unit 231 is input to the SD signal generation unit 232 and the normal equation generation unit 278.

  The HD signal is subjected to horizontal and vertical thinning processing by the SD signal generation unit 232 to generate an SD signal (525i signal) as a student signal of the learning image corresponding to the selected image number.

  In this case, the SD signal generation unit 232 generates a parameter r for determining the spatial band of the band limiting filter used when generating the SD signal from the HD signal and a parameter z for determining the noise addition state, in other words, generated. A parameter r for determining the spatial resolution of the SD signal and a parameter z for determining the amount of noise are input.

  Then, in the SD signal generation unit 232, the spatial direction of the band limiting filter used when generating the SD signal from the HD signal is changed according to the value of the parameter r, and according to the value of the input parameter r itself. The noise addition state is varied step by step.

  The parameters r and z that are input to the SD signal generator 232 are sequentially changed to change the spatial band and noise addition state of the band limiting filter used when generating the SD signal from the HD signal. As a result, a plurality of SD signals in which the spatial band and the noise addition state change stepwise are generated.

  Further, from the SD signal (525i signal) generated by the SD signal generation unit 232, the second tap selection unit 272 uses the spatial class tap data (SD pixel) located around the target position in the HD signal (1050i signal). Data) is selectively retrieved. The space class tap data (SD pixel data) selectively extracted by the second tap selection unit 272 is supplied to the space class detection unit 274. In the space class detection unit 274, ADRC processing is performed on each SD pixel data as space class tap data, and reclassification as class information of the space class (mainly class classification for waveform representation in space) is performed. A quantization code qi is obtained (see equation (1)).

  Further, from the SD signal generated by the SD signal generation unit 232, the third tap selection unit 273 selectively extracts data of the motion class tap (SD pixel data) located around the target pixel related to the HD signal. It is. The motion class tap data (SD pixel data) selectively extracted by the third tap selection unit 273 is supplied to the motion class detection unit 275. In this motion class detection unit 275, class information MV of a motion class (mainly class classification for representing the degree of motion) is obtained from each SD pixel data as motion class tap data.

  The class information MV and the requantization code qi are supplied to the first class synthesis unit 276. The first class synthesis unit 276 obtains the first class code CL1 indicating the class to which the pixel data at the target position in the HD signal (1050i signal) belongs from the supplied class information MV and the requantization code qi ( (Refer Formula (3)).

  In addition, the second class synthesis unit 277 corresponds to the selected image number input to the image set storage unit 231 and outputs the second from the motion class information MV and the space class information qi input from the learning image table 223. Get the class code CL2.

  On the other hand, from the SD signal generated by the SD signal generation unit 232, the first tap selection unit 271 selectively extracts data of predicted taps (SD pixel data) located around the target position in the HD signal.

  Then, in the normal equation generation unit 278, the class detection result of the SD signal, the first class code CL1 output from the first class synthesis unit 276, and the selection result in the learning image table 223, the second A normal equation is generated only when the second class code CL2 output from the class synthesis unit 277 matches.

  That is, in the normal equation generation unit 278, each HD pixel data y as pixel data of the target position obtained from the HD signal from the image set storage unit 231 and the first HD pixel data y corresponding to each of the HD pixel data y, respectively. Predictive tap data (SD pixel data) xi selectively extracted by the tap selection unit 271, the values of parameters r and z, and the first class composition unit 276 ( From the class code output from the second class synthesis unit 277), a normal equation (see equation (13)) for obtaining coefficient seed data w10 to wn9 is generated for each combination of class and output pixel. The

  The coefficient seed data determination unit 279 solves each normal equation to obtain coefficient seed data w10 to wn9 for each combination of class and output pixel, and these coefficient seed data w10 to wn9 are stored in the coefficient seed memory 234. Is done. The coefficient seed data is provided to the television receiver 10 via the board 12, another board, or a network (not shown).

  As described above, in the coefficient seed data generation unit 206 shown in FIG. 23, classes and output pixels (HD1 to HD4, HD1 ′ to HD4) stored in the information memory bank 61 of the image signal processing unit 35 of FIG. It is possible to generate coefficient seed data w10 to wn9 which are coefficient data in a generation expression (see Expression (5)) for obtaining coefficient data Wi used in the estimation expression for each combination of ′).

  Further, in the coefficient seed data generation unit 206, for example, when the television receiver 10 is upgraded, when generating the coefficient seed data w10 to wn9 stored in the information memory bank 61, the TV Feature amount frequency distribution similar to the feature amount frequency distribution of the user viewing image previously viewed by the user in the television receiver 10, which is stored in the history information memory 13 of the board 12 removed from the John receiver 10. Is selected, and only the selected learning image is used.

  Therefore, since the coefficient seed data can be obtained using only the selected learning image, compared to the case where the coefficient seed data is obtained from a plurality of kinds of learning images as in the prior art, the coefficient seed data can be generated efficiently. Can do.

  Further, since the coefficient seed data is obtained only from the selected learning image, the coefficient seed data w10 to wn9 are supplied to the television receiver 10 and stored in the information memory bank 61 for use. Thus, in the television receiver 10, the prediction error when the image is often converted by the user can be reduced, and the television receiver 10 has the highest performance in the installed user's house. It will be possible to demonstrate. As a result, the user can view images that he / she often views with optimum quality.

  Next, image signal processing for obtaining an HD signal from an SD signal by the image signal processing unit 35 will be described with reference to the flowchart of FIG.

  In step S11, the image signal processing unit 35 acquires SD pixel data in units of frames or fields.

  In step S12, the image signal processing unit 35 determines whether or not processing of all frames or all fields of the input SD pixel data has been completed. If it is determined in step S12 that the processing has not been completed, the image signal processing unit 35 determines whether or not a user operation signal from the remote controller 41 is input to the system controller 32 in step S13.

  When a user operation signal from the remote controller 41 is input to the system controller 32, the system controller 32 supplies the notification to the information generation control unit 65 of the image signal processing unit 35. In step S13, the system It is determined that a user operation signal has been input to the controller 32, and in step S14, the feature amount frequency distribution generation unit 66 of the image signal processing unit 35 executes history information update processing. A detailed description of this history information update processing will be described later with reference to FIG.

  As a result of the history information update process in step S14, a feature amount frequency distribution, which is image feature amount information, is obtained for each combination of motion class and space class from the SD pixel data acquired in step S11. It is recorded in the information memory 13. If it is determined in step S13 that no user operation signal is input to the system controller 32, step S14 is skipped, and the process proceeds to step S15.

  In step S <b> 15, the image signal processing unit 35 acquires image quality designation values (for example, parameters r and z) input by the user's operation of the remote controller 41, for example.

  In step S16, the coefficient generation unit 62 of the image signal processing unit 35 selects the coefficient type of each combination of the read image quality designation value, class, and output pixel (see HD1 to HD4 in FIG. 8, HD1 ′ to HD4 ′ in FIG. 9). Using the data, for example, coefficient data Wi of the estimation formula (see formula (4)) of each combination is generated by the generation formula of formula (5).

  The coefficient seed data is obtained for each combination of motion class and space class from the SD pixel data acquired in step S11 in step S14, and the feature amount frequency distribution of the image recorded in the history information memory 13 is obtained. The data is used and is generated by the prediction coefficient generation device 20.

  In step S17, the image signal processing unit 35 acquires pixel data of class taps and prediction taps corresponding to each HD pixel data to be generated from the SD pixel data acquired in step S11.

  In step S <b> 18, the image signal processing unit 35 determines whether or not the processing for obtaining HD pixel data has been completed in the entire area of the acquired SD pixel data. If it is determined in step S18 that the processing has been completed in the entire area of the acquired SD pixel data, the processing returns to step S11, and the same processing is repeated, so that the SD of the next frame or field is repeated. Processing on the pixel data is executed.

  If it is determined in step S18 that the processing has not been completed in the entire area of the acquired SD pixel data, in step S19, the class combining unit 56 of the image signal processing unit 35 acquires the class tap acquired in step S17. Class code CL is generated from the SD pixel data.

  In step S20, the image signal processing unit 35 uses the coefficient data Wi corresponding to the class code CL and the SD pixel data of the prediction tap to generate HD pixel data using an estimation formula. After the process of step S20 is complete | finished, a process returns to step S17 and the process similar to the process mentioned above is repeated.

  If it is determined in step S12 that the processing for all frames or all fields has been completed, the processing ends.

  As described above, the image signal processing unit 35 stores the feature amount frequency distribution of the user viewing image in advance, and uses the coefficient seed data generated by using the feature amount frequency distribution data of the image. By performing image processing, the television receiver 10 can generate HD image data with improved prediction accuracy for content (images) often viewed by the user.

  Therefore, the television receiver 10 can exhibit the highest performance in the installed user's house.

  Next, the history information update processing in step S14 in FIG. 24 by the feature amount frequency distribution generation unit 66 will be described in detail with reference to the flowchart in FIG.

  In step S41, the motion class detection unit 74 of the feature quantity frequency distribution generation unit 66 acquires each SD as data of the motion class tap acquired in step S11 of FIG. 24 and selectively extracted by the motion class tap selection unit 71. From the pixel data, the motion class of the SD pixel surrounded by the 4 pixels in the unit pixel block constituting the HD signal to be created is detected, and the class information (MV) of the detected motion class of the SD pixel is detected as the feature frequency distribution. The result is output to the calculation unit 77.

  In step S42, the space class detection unit 75 obtains the HD to be created from each SD pixel data as the space class tap data acquired in step S11 of FIG. 24 and selectively extracted by the space class tap selection unit 72. A requantization code (qi) is obtained as class information of the spatial class of the SD pixel surrounded by four pixels in the unit pixel block constituting the signal, and is output to the feature amount frequency distribution calculation unit 77.

  In step S43, the feature quantity generation tap selection unit 73 is created from the SD pixel data as the feature quantity generation tap data acquired in step S11 of FIG. 24 and selectively extracted by the feature quantity generation tap selection unit 73. The feature amount of the SD pixel surrounded by the four pixels in the unit pixel block constituting the HD signal to be generated is generated, and the generated feature value of the SD pixel is output to the feature amount frequency distribution calculation unit 77.

  In step S44, the feature quantity frequency distribution calculation unit 77 counts the feature quantity frequency for each combination of the supplied motion class and space class, and generates a feature quantity frequency distribution (histogram).

  In step S45, the feature quantity frequency distribution calculating unit 77 associates the feature quantity frequency distribution data generated in step S44 with the operation history information of the operation signal stored in the history information memory 13, and the motion class and Record each space class combination.

  The feature amount frequency distribution data of the user viewing image for each combination of the motion class and the space class recorded in the history information memory 13 as described above is the coefficient seed data of the television receiver 10 in the prediction coefficient generator 20. Used when updating.

  Next, the prediction coefficient generation processing of the prediction coefficient generation device 20 will be described with reference to the flowchart of FIG.

  In order to update the coefficient seed data of the television receiver 10, the board 12 on which the history information memory 13 is configured is removed from the television receiver 10 and connected to the prediction coefficient generator 20.

  In step S <b> 61, the prediction coefficient generation apparatus 20 inputs an HD signal (1050i signal) as a teaching signal of several tens of kinds of learning images via the input terminal 202, and inputs the input terminal 202 via the input terminal 201. The image number of the learning image corresponding to the teacher signal input from is input. At this time, parameters r and z are input from the input terminal 203.

  In step S <b> 62, the prediction coefficient generation device 20 receives and stores HD signals as teaching signals of dozens of types of learning images from the input terminal 202 in the image set storage unit 231 of the coefficient seed data generation unit 206.

  In step S <b> 63, the learning image frequency distribution generation unit 204 of the prediction coefficient generation device 20 receives the HD signal and the image number input from the input terminals 201 and 202, and executes a learning image feature frequency distribution generation process. . The learning image feature amount frequency distribution generation processing will be described later in detail with reference to FIG.

  In the learning image feature amount frequency distribution generation processing in step S63, a feature amount frequency distribution of the learning image is generated for each combination of the motion class and the space class, and the data is associated with the image number of the learning image, and the learning image It is recorded in the frequency distribution memory 213.

  In step S64, the prediction coefficient generation device 20 determines whether or not all learning images have been input. If it is determined that all learning images have not been input, the process returns to step S61, and thereafter The process is repeated.

  If it is determined in step S64 that all learning images have been input, in step S65, the prediction coefficient generation device 20 selects one combination of motion class and space class.

  In step S66, the learning image selection unit 205 of the prediction coefficient generation device 20 performs learning image selection processing for the combination of the motion class and the space class selected in step S65. This learning image selection processing will be described later in detail with reference to FIG.

  With the learning image selection processing in step S66, the feature amount frequency distribution of the user viewing image stored in the history information memory 13 of the board 12 and the learning image frequency distribution memory 213 for one combination of motion class and space class are stored. The feature amount frequency distributions of all recorded learning images are compared, and a learning image similar to the user viewing image is selected. Then, the image number of the selected learning image is recorded in the learning image table 223 together with the combination of the motion class and the space class as the selected image number.

  In step S67, the coefficient seed data generation unit 206 executes coefficient seed data generation processing of the combination of the corresponding motion class and space class using the learning image of the selected image number recorded in the learning image table 223 in step S66. To do. The coefficient seed data generation process will be described later in detail with reference to FIG.

  By the coefficient seed data generation processing in step S67, coefficient seed data of a combination of one motion class and a space class is generated and generated using only the learning image that is similar to the user viewing image in step S66. Coefficient seed data is recorded.

  In step S68, the prediction coefficient generation device 20 determines whether coefficient seed data for all the motion class and space class combinations has been obtained, and has yet to obtain coefficient seed data for all the motion class and space class combinations. If it is determined that there is no such process, the process returns to step S65, and the subsequent processes are repeated.

  If it is determined in step S68 that coefficient seed data for all combinations of motion classes and space classes has been obtained, the process ends.

  Next, the learning image feature amount frequency distribution generation processing in step S63 in FIG. 26 will be described in detail with reference to the flowchart in FIG. Note that the processing in steps S82 to S85 in FIG. 27 is basically the same as the processing in steps S41 to S44 in FIG.

  In step S <b> 81, the SD signal generation unit 211 performs horizontal and vertical thinning processing on the learning image HD signal from the input terminal 202 using the values of the parameters r and z input from the input terminal 203. To obtain an SD signal as a student signal of the learning image.

  In step S82, the motion class detection unit 254 of the feature amount frequency distribution generation unit 212 is obtained from each SD pixel data as the motion class tap data obtained in step S81 and selectively extracted by the motion class tap selection unit 251. The motion class of the SD pixel surrounded by the four pixels in the unit pixel block constituting the HD signal to be created is detected, and the class information (MV) of the detected motion class of the SD pixel is detected as the feature amount frequency distribution calculation unit 257. Output to.

  In step S83, the space class detection unit 255 configures an HD signal to be generated from each SD pixel data as the space class tap data acquired in step S81 and selectively extracted by the space class tap selection unit 252. The requantization code (qi) as class information of the spatial class of the SD pixel surrounded by the four pixels in the unit pixel block to be obtained is obtained and output to the feature amount frequency distribution calculation unit 257.

  In step S84, the feature amount generation tap selection unit 253 obtains the HD to be created from the SD pixel data as the feature amount generation tap data acquired in step S81 and selectively extracted by the feature amount generation tap selection unit 253. The feature amount of the SD pixel surrounded by the four pixels in the unit pixel block constituting the signal is generated, and the generated feature value of the SD pixel is output to the feature amount frequency distribution calculation unit 257.

  In step S85, the feature amount frequency distribution calculation unit 257 counts the feature amount frequency for each combination of the supplied motion class and space class, and generates a feature amount frequency distribution (histogram).

  In step S86, the feature quantity frequency distribution calculation unit 257 associates the data of the feature quantity frequency distribution generated in step S85 with the image number received from the input terminal 201 in step S61 in FIG. Each class combination is recorded in the learning image frequency distribution memory 213. Thereby, the feature amount frequency distribution generation process is terminated, and the process returns to step S63 in FIG.

  Next, the learning image selection processing in step S66 in FIG. 26 will be described with reference to the flowchart in FIG.

  In step S101, the frequency distribution comparison unit 221 uses the history information memory 12 and the learning image frequency distribution for the feature amount frequency distribution data of all the images for the combination of the motion class and the space class selected in step S65 of FIG. Read from the memory 213.

  At this time, the frequency distribution comparison unit 221 averages and collects the feature amount frequency distributions of the user viewing images from the history information memory 12 into one. Although it is possible to compare one image at a time without averaging, the process of averaging is more efficient with less calculation load.

  In step S <b> 102, the frequency distribution comparison unit 221 compares the feature amount frequency distribution of the user viewing image with the feature amount frequency distribution of all the learning images in the learning image frequency distribution memory 213.

  In step S103, the learning image selection unit 222 selects a learning image that most closely resembles the feature amount frequency distribution of the user viewing image based on the comparison result in step S102, and associates it with the combination of the motion class and the space class. The selected image number is recorded in the learning table 223. Thereby, the image selection process is ended, and the process returns to step S66 in FIG.

  Next, the coefficient seed data generation process in step S67 of FIG. 26 will be described in detail with reference to the flowchart of FIG.

  The information on the motion class and the space class recorded in the learning table 223 in step S103 of FIG. 28 is input to the second class synthesis unit 277, and the selected image number corresponding to the motion class and the space class is stored in the learning set storage. Is input to the unit 231.

  In addition, the learning set storage unit 231 stores HD signals of several tens of types of learning images in step S62 in FIG.

  In step S121, the coefficient seed data generation unit 206 receives the selection image number from the learning image table 223 in the learning set storage unit 231, and selects the HD signal of the corresponding learning image.

  In step S122, the coefficient seed data generation unit 206 acquires HD image data from the learning set storage unit 231.

  In step S123, the coefficient seed data generation unit 206 determines whether or not processing has been completed for all HD pixel data. If it is determined in step S123 that the process has not been completed for all HD pixel data, the process proceeds to step S124.

  In step S124, the SD signal generation unit 232 of the coefficient seed data generation unit 206 performs horizontal and vertical thinning processing on the HD pixel data acquired in step S122 to generate SD pixel data.

  In step S125, the coefficient seed data generation unit 206 performs steps from the SD pixel data generated in step S124 in the first tap selection unit 271, the second tap selection unit 272, and the third tap selection unit 273. Corresponding to each HD pixel data acquired in S122, pixel data of class tap and prediction tap is acquired.

  In step S126, the coefficient seed data generation unit 206 determines whether or not the learning process has been completed in the entire region of the generated SD pixel data. If it is determined in step S126 that the learning process has been completed, the process returns to step S122, the next HD pixel data is acquired, and the same process is repeated.

  If it is determined in step S126 that the learning process has not ended, the first class synthesis unit 276 of the coefficient seed data generation unit 206 determines from the SD pixel data of the class tap acquired in step S125 in step S127. A first class code is generated.

  In step S128, the second class synthesis unit 277 of the coefficient seed data generation unit 206 generates a second class code using the motion class and space class information from the learning table 223.

  In step S129, the normal equation generation unit 278 of the coefficient seed data generation unit 206 generates a normal equation (see equation (13)) when the first class code and the second class code match. When the process of step S129 ends, the coefficient seed data generation unit 278 returns the process to step S125 and repeats the subsequent processes.

  If it is determined in step S123 that learning has been completed for all image quality patterns, the coefficient seed data determination unit 279 of the coefficient seed data generation unit 206 solves the normal equation by a sweep-out method or the like in step S130. The coefficient seed data of each combination of class and output pixel (see HD1 to HD4 in FIG. 8, HD1 ′ to HD4 ′ in FIG. 9) is calculated.

  In step S 131, the coefficient seed data determination unit 279 of the coefficient seed data generation unit 206 records the coefficient seed data in the coefficient seed memory 234. When the process of step S131 ends, the coefficient seed data generation process ends, and the process returns to step S67 of FIG.

  As described above, the prediction coefficient generation device 20 has the most similar shape to the feature amount frequency distribution of the user viewing image stored in the history information memory 13 of the board 12 taken out from the television receiving device 10. Coefficient seed data used in the television receiver 10 is generated using only the learning image having the feature quantity frequency distribution.

  Thereby, in the television receiver 10 installed in the user's home, the prediction error of the conversion is reduced and the prediction accuracy is improved for the content image that the user often views. You can see the image converted to image quality. That is, the television receiver 10 can exhibit the best performance at the user's home.

  As described above, in the image signal processing system shown in FIG. 4, the television receiver 10 can attach / detach the feature frequency distribution data of the viewing image of the user, which is an image of content often viewed by the user. When the coefficient seed data recorded in the history information memory 13 provided in the substrate 12 and used in the television receiver 10 is updated, the prediction coefficient generation device 20 of the substrate 12 removed from the television receiver 10 is used. The coefficient seed data is generated using a learning image having the most similar feature frequency distribution to the user viewing image recorded in the history information memory 13.

  Thereby, in the prediction coefficient production | generation apparatus 20, the coefficient seed data for making the television receiver 10 exhibit the highest performance can be produced | generated efficiently. Furthermore, by using the coefficient seed data, the television receiver 10 can convert and display an image of content viewed by the user so as to obtain an optimum quality with little prediction error.

  In the image signal processing unit 35 in FIG. 4, the generation formula of the formula (5) is used to generate the coefficient data Wi (i = 1 to n), but a polynomial having a different order or another function is used. The generation of the coefficient data Wi can be realized even by the expressed expression.

  The image signal processing unit 35 sets a parameter r that determines the resolution in the spatial direction (vertical direction and horizontal direction) and a parameter z that determines the degree of noise removal, and adjusts the values of these parameters r and z. Although it has been described that the resolution in the spatial direction of an image and the degree of noise removal can be adjusted, other devices that provide parameters for determining the quality of an image can be similarly configured. For example, the parameters include a parameter that determines the resolution in the vertical direction, a parameter that determines the resolution in the horizontal direction, a parameter that determines the resolution in the time direction (frame direction), and the degree of change in the phase of the pixel after conversion (vertical direction and horizontal direction) Various parameters such as a parameter for determining the brightness of the screen, a parameter for determining the contrast of the screen, and a parameter for determining the playback speed are conceivable.

  The values of these parameters may be directly specified by an operation input by the user, or parameter values related to the editing operation content are automatically set based on the content edited by the user. You may do it.

  For example, when the user instructs to display a zoom image by operating the remote controller 41 or the like, in the zoom image enlargement ratio, change speed, or original image at the center of the zoom image specified by the user at that time Based on the information such as the position, parameters such as vertical and horizontal phase change information of the pixel, the resolution of the zoom image, or the noise removal degree of the zoom image, which are necessary when creating the zoom image, are generated. You may do it.

  Further, when the user instructs to perform slow motion playback by operating the remote controller 41 or the like, not only the playback speed of the image but also the image playback speed is specified based on information such as the playback speed specified by the user at that time. Parameters such as resolution and noise removal degree may be generated.

  Further, an illuminance meter or the like is installed in the television receiver 10 to measure the brightness around the television receiver 10 and parameters such as the brightness of the image are generated based on the measurement result. For example, the type of display of the television receiver 10 that displays an image (CRT (Cathode Ray Tube), LCD (Liquid Crystal Display), or PDP (Plasma Display Panel)), information such as the screen size Based on the above, parameters such as image resolution, noise removal, brightness, or contrast may be generated.

  Further, although the image signal processing unit 35 has been described as being capable of adjusting two parameters r and z, one that handles one or three or more parameters can be similarly configured.

  Furthermore, although the prediction coefficient generation device 20 has been described as generating coefficient seed data using the generation method of FIG. 16, other methods can be used as the coefficient seed data generation method. In the generation method of FIG. 16, learning is performed between a plurality of SD signals generated by varying the parameters and the HD signal to generate coefficient seed data. What is the generation method? Differently, for example, learning is performed for each of a plurality of SD signals generated by changing parameters with an HD signal, and coefficient data Wi of the estimation formula of Formula (4) is generated to correspond to each SD signal. A method of generating coefficient seed data using the coefficient data Wi generated as described above can also be used. Note that coefficient data can be obtained instead of coefficient seed data.

  Further, in the above description, when comparing the feature amount frequency distribution of the user viewing image and the feature amount frequency distribution of the learning image, the case where the correlation coefficient C is used as a comparison value has been described. Other values can be used as long as the values are not limited to the correlation coefficient C and can be used to evaluate whether both pieces of feature amount information are similar or different. In this case, the case where the comparison value is large is not necessarily evaluated as being similar, but is evaluated according to the type of the comparison value.

  In the above description, the description has been given using the television receiver having the tuner and the monitor that converts the SD signal into the HD signal. However, the present invention is not limited to the television receiver, and the SD signal is converted into the HD signal. Any device that converts a signal can be applied to an image processing device, for example.

  The series of processes described above can be executed by hardware or can be executed by software.

  When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer incorporated in dedicated hardware or various programs. For example, it is installed from a program recording medium in a general-purpose personal computer or the like.

  FIG. 30 is a block diagram illustrating an example of a configuration of a personal computer 401 that executes the above-described series of processing by a program. A CPU (Central Processing Unit) 411 executes various processes according to a program stored in a ROM (Read Only Memory) 412 or a storage unit 418. A RAM (Random Access Memory) 413 appropriately stores programs executed by the CPU 411 and data. The CPU 411, the ROM 412, and the RAM 413 are connected to each other by a bus 414.

  An input / output interface 415 is also connected to the CPU 411 via the bus 414. The input / output interface 415 is connected to an input unit 416 including a keyboard, a mouse, and a microphone, and an output unit 417 including a display and a speaker. The CPU 411 executes various processes in response to commands input from the input unit 416. Then, the CPU 411 outputs the processing result to the output unit 417.

  The storage unit 418 connected to the input / output interface 415 includes, for example, a hard disk, and stores programs executed by the CPU 411 and various data. The communication unit 419 communicates with an external device via a network such as the Internet or a local area network.

  A program may be acquired via the communication unit 419 and stored in the storage unit 418.

  The drive 420 connected to the input / output interface 415 drives a removable medium 421 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, and drives the programs and data recorded therein. Get etc. The acquired program and data are transferred to and stored in the storage unit 418 as necessary.

  As shown in FIG. 30, a program recording medium for storing a program that is installed in a computer and is ready to be executed by the computer includes a magnetic disk (including a flexible disk), an optical disk (CD-ROM (Compact Disc-Read Only). Memory, DVD (Digital Versatile Disc), a magneto-optical disk, a removable medium 421 that is a package medium composed of a semiconductor memory, a ROM 412 in which a program is temporarily or permanently stored, and a storage unit 418 It is comprised by the hard disk etc. which comprise. The program is stored in the program recording medium using a wired or wireless communication medium such as a local area network, the Internet, or digital satellite broadcasting via a communication unit 419 that is an interface such as a router or a modem as necessary. Done.

  In the present specification, the step of describing the program stored in the program recording medium is not limited to the processing performed in time series in the order described, but is not necessarily performed in time series. Or the process performed separately is also included.

  Further, in this specification, the system represents the entire apparatus constituted by a plurality of apparatuses.

It is a figure explaining the outline | summary of this invention. It is a figure explaining the outline | summary of this invention. It is a figure explaining the outline | summary of this invention. It is a figure which shows the structure of one Embodiment of the image signal processing system to which this invention is applied. It is a block diagram which shows the structural example of the television receiver of FIG. It is a figure which shows the pixel positional relationship of a 525i signal and a 1050i signal. It is a block diagram which shows the structural example of the image signal processing part of FIG. It is a figure which shows the phase shift (odd field) from the center prediction tap of 4 pixels in the unit pixel block of HD signal (1050i signal). It is a figure which shows the phase shift (even field) from the center prediction tap of 4 pixels in the unit pixel block of HD signal (1050i signal). It is a block diagram which shows the structural example of the feature-value frequency distribution generation part of FIG. It is a figure explaining the example of a feature-value It is a figure which shows the example of the feature-value value of a pixel, and feature-value frequency distribution. It is a figure which shows the example of the feature-value frequency distribution memorize | stored in the log | history information memory of FIG. It is a figure which shows the example of a user interface for adjusting an image quality. It is the figure which expanded and showed the adjustment screen of FIG. It is a figure which shows an example of the production | generation method of coefficient seed data. It is a block diagram which shows the structural example of the prediction coefficient production | generation apparatus of FIG. It is a block diagram which shows the structural example of the feature-value frequency distribution generation part of FIG. It is a figure which shows the example of the feature-value frequency distribution memorize | stored in the learning image frequency distribution memory of FIG. It is a figure explaining the comparison object of the feature-value frequency distribution by the frequency distribution comparison part of FIG. It is a figure which shows the example of the comparison method of the feature-value frequency distribution by the frequency distribution comparison part of FIG. It is a figure which shows the structural example of the learning image selection of the learning image selection part of FIG. 17, and a learning image table. It is a figure which shows the detailed structural example of the prediction coefficient learning part of FIG. 5 is a flowchart for explaining image signal processing by the television receiver of FIG. 4. It is a flowchart explaining the update process of the historical information of step S14 of FIG. It is a flowchart explaining the prediction coefficient production | generation process by the prediction coefficient production | generation apparatus of FIG. It is a flowchart explaining the feature-value frequency distribution generation process of the learning image of step S63 of FIG. It is a flowchart explaining the learning image selection process of step S66 of FIG. It is a flowchart explaining the coefficient seed data generation process of step S67 of FIG. It is a block diagram which shows the structural example of the personal computer to which this invention is applied.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 Television receiver, 12 Board | substrate, 13, History information memory, 20 Prediction coefficient production | generation apparatus, 21 Coefficient seed | species data production | generation part, 32 System controller, 35 Image signal processing part, 65 Information production | generation control part, 66 Feature-value frequency distribution production | generation , 73 feature quantity generation tap selection section, 76 feature quantity generation section, 77 feature quantity frequency distribution calculation section, 204 learning image frequency distribution generation section, 205 learning image selection section, 206 coefficient seed data generation section, 207 system controller, 211 SD signal generation unit, 212 feature frequency distribution generation unit, 213 learning image frequency distribution memory, 221 frequency distribution comparison unit, 222 learning image selection unit, 223 learning image table, 231 image set storage unit, 232 SD signal generation unit, 233 prediction Coefficient learning unit, 234 Coefficient seed method Li, 253 feature quantity generating tap selector, 256 feature quantity generation unit, 257 feature quantity the frequency distribution calculation section, 276 first class synthesis unit, 277 second class synthesis unit

Claims (13)

In a signal processing device that generates a coefficient used for an operation for converting a first image signal into a second image signal having a higher resolution than the first image signal,
From feature amount storage means for storing feature amount information of an image corresponding to the first image signal generated by another signal processing device that converts the first image signal into the second image signal, Obtaining means for obtaining feature amount information of the image;
A signal processing apparatus comprising: coefficient generation means for generating a coefficient used for an operation for conversion to the second image signal by the other signal processing apparatus using the feature amount information of the image acquired by the acquisition means. . The signal processing apparatus according to claim 1, wherein the feature amount storage unit is provided on a substrate that can be attached to and detached from the other signal processing apparatus. The signal processing device according to claim 1, wherein the feature amount information of the image is a feature amount frequency distribution of the image. Feature quantity generation means for generating feature quantity information of the learning image;
An image for selecting an image similar to the image from the learning image using the feature amount information of the learning image generated by the feature amount generation unit and the feature amount information of the image acquired by the acquisition unit And a selection means,
The coefficient generation unit generates a coefficient used for an operation of converting the learning image selected by the image selection unit into the second image signal by the other signal processing device. Signal processing equipment. In a signal processing method of a signal processing device that generates a coefficient used for an operation for converting a first image signal into a second image signal having a higher resolution than the first image signal,
From feature amount storage means for storing feature amount information of an image corresponding to the first image signal generated by another signal processing device that converts the first image signal into the second image signal, Obtaining feature amount information of the image;
A signal processing method including a step of generating a coefficient used for an operation of converting the acquired image feature quantity information into the second image signal by the other signal processing device. A program for causing a computer to perform a process of generating a coefficient used for an operation for converting a first image signal into a second image signal having a higher resolution than the first image signal,
From the feature amount storage means for storing feature amount information of an image corresponding to the first image signal generated by a signal processing device that converts the first image signal into the second image signal, the image Feature amount information,
A program including a step of generating a coefficient used for an operation of converting the acquired image feature quantity information into the second image signal by the signal processing device.   A recording medium on which the program according to claim 6 is recorded. In the signal processing device that converts the first image signal into a second image signal having a higher resolution than the first image signal,
Feature amount information generating means for generating feature amount information of an image corresponding to the first image signal to be converted into the second image signal;
Feature quantity storage means for storing feature quantity information of the image generated by the feature quantity information generation means;
The first image signal is converted into the second image signal by performing calculation using a coefficient generated by another signal processing device using the image feature value information stored in the feature value storage means. A signal processing device comprising: an image signal converting means for converting to an image signal. The signal processing apparatus according to claim 8, wherein the feature amount storage unit is provided on a detachable substrate. The signal processing apparatus according to claim 8, wherein the image feature amount information is a feature amount frequency distribution of the image. In the signal processing method of the signal processing device for converting the first image signal into the second image signal having a higher resolution than the first image signal,
Generating feature amount information of an image corresponding to the first image signal to be converted into the second image signal;
Storing the generated feature amount information of the image in the feature amount storage means;
The first image signal is converted into the second image signal by performing calculation using a coefficient generated by another signal processing device using the image feature value information stored in the feature value storage means. A signal processing method including a step of converting into an image signal. A program for causing a computer to perform a process of converting a first image signal into a second image signal having a higher resolution than the first image signal,
Generating feature amount information of an image corresponding to the first image signal to be converted into the second image signal;
The feature amount information of the generated image is stored in the feature amount storage means,
The first image signal is converted into the second image signal by performing calculation using the coefficient generated by the signal processing device using the feature amount information of the image stored in the feature amount storage unit. A program that includes steps to convert to.   A recording medium on which the program according to claim 12 is recorded.

Images