JP4093114B2 - Information signal processing apparatus and processing method, codebook generation apparatus and generation method, and program for executing each method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention can be applied to, for example, a standard television signal (SD signal) corresponding to a standard or low resolution into a high resolution signal (HD signal), an information signal processing apparatus, a processing method, and a code suitable for use. The present invention relates to a book generation apparatus and generation method, and a program for executing each method.
[0002]
Specifically, the present invention provides a first information signal comprising a plurality of information data. More sample pixels than the first information signal It is obtained by dividing the first information signal when converting to the second information signal composed of a plurality of information data. Double For each block consisting of several pieces of information data, the class tap class code consisting of the information data contained in this block and Consisting of dynamic range Feature value Using Configure the second information signal Double By generating several pieces of information data, the first information signal is High resolution The present invention relates to an information signal processing apparatus or the like that can be favorably converted into a second information signal.
[0003]
[Prior art]
For example, there is an image signal processing method for classifying an image signal into one of a plurality of classes and performing processing corresponding to the class (Patent Document 1 and Patent Document 2). reference).
[0004]
According to such an image signal processing method, for example, when class classification is performed based on the activity of the image signal, an image signal with a high activity such as a rapidly changing image signal and an activity such as a flat image signal Image signals with low activity can be classified into different classes, and as a result, it is possible to perform processing suitable for image signals with high activity and image signals with low activity.
[0005]
[Patent Document 1]
JP-A-7-95591
[Patent Document 2]
JP 2000-59740 A
[0006]
[Problems to be solved by the invention]
According to the image signal processing for class classification as described above, image signal processing suitable for the image signal of the class can be performed for each image signal classified into each class. Therefore, theoretically, the larger the number of classes, the more suitable processing can be performed on the image signals classified into each class. However, when the number of classes becomes enormous, the pattern of processing performed according to the class becomes enormous, which means that the apparatus becomes larger.
[0007]
For example, as described above, when classifying based on an activity, the same class as the number of values that the activity can take is prepared, and the image signal of each activity is assigned to the activity. Appropriate treatment can be performed. However, as an activity, for example, when N difference values between adjacent pixels arranged in a horizontal direction are adopted, if the difference value is K bits, the total number of classes is (2 ^K ) ^N It becomes a huge number of classes.
[0008]
For this reason, the classification is performed using some kind of compression processing such as ADRC (Adaptive Dynamic Range Coding). In class classification using ADRC, for example, as described above, N difference values as data used for class classification (hereinafter referred to as “class tap” as appropriate) are subjected to ADRC processing and obtained as a result. Class determination is performed according to the ADRC code.
[0009]
In the K-bit ADRC, for example, the maximum value MAX and the minimum value MIN of the data values constituting the class tap are detected, and DR = MAX−MIN is set as the local dynamic range of the set, and this dynamic range DR The minimum value MIN is subtracted from the data values constituting the class tap based on ^K Divide by (quantize). Then, a bit string in which the K bit values for the data constituting the class tap obtained in the above manner are arranged in a predetermined order is output as an ADRC code.
[0010]
Therefore, when a class tap is subjected to, for example, 1-bit ADRC processing, each data constituting the class tap is divided by (MAX−MIN) / 2 after the minimum value MIN is subtracted. Each data is 1 bit. A bit string in which the 1-bit data is arranged in a predetermined order is output as an ADRC code.
[0011]
Here, the classification can be performed by, for example, vector quantization in addition to the ADRC process. As described above, when class classification is performed using compression processing, the number of classes can be reduced. On the other hand, however, it may not be possible to perform fine classification as compared with the case where classification is performed without performing compression processing, and appropriate processing may not be performed on the image signal.
[0012]
For example, consider image signal processing that classifies a first image signal using vector quantization and generates a second image signal for each class.
[0013]
In this case, the first image signal is divided into a plurality of blocks of pixel data, and for each block, a vector having a plurality of pixel data constituting the block as components (hereinafter, referred to as “block vector” as appropriate). Composed. Further, the block vector is subjected to vector quantization using a code book obtained in advance, and a code (symbol) as the vector quantization result is output as a class code representing the class of the block vector.
[0014]
Then, pixel data of the second image signal is generated for each class code. That is, the class code is subjected to vector inverse quantization using the code book, and a block vector corresponding to the class code is obtained. A block including the component of the block vector as pixel data is obtained, and the second image signal is generated by arranging the block at a corresponding position.
[0015]
However, if the second image signal obtained by vector inverse quantization is called a high-quality image signal and the first image signal to be vector-quantized is called a low-quality image signal, In the processing, all the block vectors of the low-quality image signal that are vector-quantized to the same class code are vector-inverted to the same code vector, that is, the block vector of the same high-quality image signal. Quantized. That is, a block vector obtained by vector inverse quantization has a so-called quantization error. Therefore, even though the image signal generated by the above-described processing has a high image quality, if the image quality of the high-quality image signal used for creating the code book for vector inverse quantization is used, the quantization error component Only the image quality is degraded.
[0016]
An object of the present invention is to satisfactorily convert a first information signal into a second information signal.
[0017]
[Means for Solving the Problems]
The information signal processing apparatus according to the present invention converts a first information signal made up of a plurality of information data into a second information signal made up of a plurality of information data having a larger number of sample pixels than the first information signal. An information signal processing device for conversion, wherein the first information signal is divided into blocks each composed of a plurality of information data, and for each block, class tap extraction means for extracting the information data contained in the block as a class tap. , The class tap extracted by this class tap extraction means , Based on the signal value of each tap Classify into one of several classes , Representing the characteristics of the block A class classification unit that obtains a class code, and a feature that obtains a feature amount related to the class tap by obtaining a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the class tap extracted by the class tap extraction unit The representative vector assigned to the combination of the feature quantity consisting of the class code obtained by the quantity acquisition means and the class classification means and the dynamic range obtained by the feature quantity acquisition means is obtained from the data generation codebook, and is used as a class tap. Corresponding data generating means for generating a plurality of pieces of information data constituting the second information signal. The class classification means vector-quantizes a tap vector having information data included in the class tap extracted by the class tap extraction means as a component using a class classification codebook, and class corresponding to the class tap. Get the code Is.
[0018]
In the information signal processing method according to the present invention, the first information signal composed of a plurality of pieces of information data is converted into a second information composed of a plurality of pieces of information data having a larger number of sample pixels than the first information signal. An information signal processing method for converting to a signal, wherein the first information signal is divided into blocks each composed of a plurality of information data, and for each block, the information data included in the block is extracted as a class tap. Process and the class tap extracted in this class tap extraction process , Based on the signal value of each tap Classify into one of several classes , Representing the characteristics of the block A feature for obtaining a feature amount related to a class tap by obtaining a dynamic range that is a difference between a maximum value and a minimum value of pixel data in the class tap extracted in the class tap extraction step and a class classification step for obtaining a class code The representative vector assigned to the combination of the feature quantity consisting of the class code obtained in the quantity acquisition process, the class classification process, and the dynamic range obtained in the feature quantity acquisition process is obtained from the data generation codebook, and is assigned to the class tap. A corresponding data generation step for generating a plurality of information data constituting the second information signal. In the class classification process, the tap vector having the information data included in the class tap extracted in the class tap extraction process as a component is vector quantized using the class classification codebook, and the class tap is supported. Includes processing to get the class code Is.
[0019]
A program according to the present invention is for causing a computer to execute the above information signal processing method.
[0020]
In the present invention, a first information signal composed of a plurality of information data is converted into a second information signal composed of a plurality of information data. Here, the information signal is, for example, an image signal composed of a plurality of pixel data (sample data) or an audio signal composed of a plurality of sample data.
[0021]
A 1st information signal is divided | segmented into the block which consists of several information data, and the information data contained in this block are extracted as a class tap for every block. And this class tap , Based on the signal value of each tap Classified into one of several classes , Representing the characteristics of the block A class code is obtained. For example, a tap vector having information data included in a class tap as a component is vector quantized using a data generation code book, and a class code corresponding to the class tap is obtained.
[0022]
Further, based on the class tap, the feature amount related to the class tap is acquired. This feature amount is the dynamic range, variance value, standard deviation, waveform pattern, or difference data between the tap vector and the code vector corresponding to the class code.
[0023]
Then, as described above, the class code and feature amount obtained based on the class tap Using, Configure the second information signal corresponding to the class tap Ru A plurality of information data is generated. For example ,this The class code is For data generation The second information signal is dequantized using the code book to form the second information signal. Double A code vector having several pieces of information data as components is obtained. This code book is obtained by learning in advance.
[0024]
Thus, the present invention provides a first information signal comprising a plurality of information data. More sample pixels than the first information signal It is obtained by dividing the first information signal when converting to the second information signal composed of a plurality of information data. Double For each block consisting of several pieces of information data, the class tap class code consisting of the information data contained in this block and Consisting of dynamic range Feature value The representative vector assigned to the combination of is obtained from the data generation codebook, Configure the second information signal Double Since several pieces of information data are generated, it is possible to perform an appropriate conversion process as compared with the case based on only the class code, and to convert the first information signal into the second information signal satisfactorily.
[0025]
In the code book generating device according to the present invention, the first information signal composed of a plurality of information data is converted into a second information composed of a plurality of information data having a larger number of sample pixels than the first information signal. A class book of class taps composed of information data included in a block for each block composed of a plurality of information data obtained by dividing the first information signal, which is a code book used for conversion into a signal Dividing a student signal corresponding to a first information signal in a codebook generation device for generating a data generation codebook in which a code vector as a representative vector for a second information signal is assigned to a combination of a code and a feature amount The information data contained in each first block made up of a plurality of information data obtained as a first tap is extracted as a first tap. The information data included in each second block made up of a plurality of pieces of information data obtained by dividing the loop extraction means and the teacher signal corresponding to the second information signal corresponding to the first block A second tap extracting means for extracting the first tap extracted by the first tap extracting means. , Based on the signal value of each tap Classify into one of several classes , Representing the characteristics of the block A class classification unit that obtains a class code, and a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the first tap extracted by the first tap extraction unit is obtained to relate to the first tap. A plurality of feature data acquisition means for acquiring a feature quantity and a plurality of information data components corresponding to the second taps extracted by the second tap extraction means for the second information signal, respectively; Storage means for storing representative vectors; class code obtained by class classification means; feature quantity consisting of dynamic range obtained by feature quantity acquisition means; second tap extracted by second tap extraction means; and storage One selected from a plurality of representative vectors for each combination of class code and feature quantity based on a plurality of representative vectors stored in the means A codebook generating means for generating data generating codebooks assign a representative vector The class classification means vector-quantizes the tap vector having the information data contained in the first tap extracted by the first class tap extraction means as a component using a class classification codebook, Get the class code corresponding to the tap Is.
[0026]
In the code book generating method according to the present invention, the first information signal composed of a plurality of information data is converted into a second information composed of a plurality of information data having a larger number of sample pixels than the first information signal. A class book of class taps composed of information data included in a block for each block composed of a plurality of information data obtained by dividing the first information signal, which is a code book used for conversion into a signal Dividing a student signal corresponding to a first information signal in a codebook generation method for generating a databook codebook in which a code vector as a representative vector for a second information signal is assigned to a combination of a code and a feature amount A first tag for extracting, as a first tap, information data included in each first block composed of a plurality of information data obtained in the above manner. Information data contained in each second block consisting of a plurality of information data obtained by dividing the teacher signal corresponding to the second information signal and the teacher signal corresponding to the first block. A second tap extraction step for extracting the first tap, and the first tap extracted in the first tap extraction step. , Based on the signal value of each tap Classify into one of several classes , Representing the characteristics of the block A class classification step for obtaining a class code, and a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the first tap extracted in the first tap extraction step is obtained to relate to the first tap. Feature quantity acquisition step for acquiring feature quantity, class code obtained in class classification process, feature quantity consisting of dynamic range obtained in feature quantity acquisition process, second tap extracted in second tap extraction process And a plurality of representative vectors having a plurality of pieces of information data as components corresponding to the second tap extracted in the second tap extraction step, respectively, for the second information signal, A code that generates a data generation codebook by assigning one representative vector selected from a plurality of representative vectors to each combination of feature quantities Yes and click generating step In the class classification step, a tap vector having the information data included in the first tap extracted in the first class tap extraction step as a component is vector quantized using a class classification codebook, Includes a process to get the class code corresponding to the first tap Is.
[0027]
A program according to the present invention is for causing a computer to execute the above-described code book generation method.
[0028]
In the present invention, it is used when converting the first information signal into the second information signal. For data generation Generate a code book. Here, the conversion from the first information signal to the second information signal is obtained by dividing the first information signal. Double For each block consisting of several pieces of information data, the class tap class code consisting of the information data contained in this block and Consisting of dynamic range Feature value Using Configure the second information signal Double This is done by obtaining a representative vector whose components are several pieces of information data.
[0029]
Obtained by dividing the student signal corresponding to the first information signal Double Information data included in each first block made up of several pieces of information data is extracted as a first tap. Also, the teacher signal corresponding to the second information signal is divided corresponding to the first block. By Obtained Double Information data included in each second block made up of several pieces of information data is extracted as a second tap.
[0030]
And this first tap , Based on the signal value of each tap Classify into one of several classes , Representing the characteristics of the block A class code is obtained. For example, a tap vector having information data included in the first tap as a component is subjected to vector quantization using a class classification code book, and a class code corresponding to the first tap is obtained.
[0031]
The first tap The dynamic range that is the difference between the maximum value and the minimum value of the pixel data in A feature amount related to one tap is acquired. This feature amount is the dynamic range of the information data in the first tap. And The variance value, standard deviation, waveform pattern, or difference data between the tap vector and the code vector corresponding to the class code are used.
[0032]
As described above, the class code and the feature amount obtained based on the first tap, the second tap, and the second information signal respectively correspond to the second tap. Double Multiple representative vectors whose components are several pieces of information data Using By assigning one representative vector selected from a plurality of representative vectors to each combination of class code and feature quantity, For data generation A codebook is generated.
[0033]
For example, for each combination of a class code and a feature amount, information data included in the second tap and a representative vector for each of a plurality of second taps corresponding to the plurality of first taps corresponding to the combination. An error with the component is obtained, and the obtained plurality of errors are added to obtain an error sum, which is repeated for a plurality of representative vectors, and a representative vector that minimizes the error sum is assigned to the combination. .
[0034]
Thus, according to the present invention, the first information signal is Many pixels To the second information signal A codebook used for conversion, Obtained by dividing the first information signal Double For each block consisting of several pieces of information data, a class code and Consisting of dynamic range Feature value For data generation in which a code vector as a representative vector for the second information signal is assigned to the combination of Good codebook generation.
[0035]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 shows a configuration of an image signal processing apparatus 100 as an embodiment. The image signal processing apparatus 100 converts a low resolution or standard resolution image signal (hereinafter referred to as “SD (Standard Definition) signal”) into a high resolution image signal (hereinafter referred to as “HD (High Definition) signal”).
[0036]
FIG. 2 shows a pixel position relationship between the SD signal and the HD signal. “X” represents the pixel position of the SD signal, and “◯” represents the pixel position of the HD signal. In this case, four pixels of the HD signal correspond to two pixels of the SD signal.
[0037]
Returning to FIG. 1, the image signal processing apparatus 100 includes a tap extraction unit 101. The tap extraction unit 101 is supplied with an SD signal as a conversion target. This SD signal is composed of a plurality of 8-bit pixel data. The tap extraction unit 101 is an SD signal Duplicate In this embodiment, the pixel data is divided into blocks composed of two pieces of pixel data p1 and p2 adjacent in the horizontal direction (see FIG. 2), and the pixel data p1 and p2 included in this block are classified into classes for each block. Extract as a tap.
[0038]
In addition, the image signal processing apparatus 100 includes a class classification unit 102. The class classification unit 102 is a cluster cluster extracted by the tap extraction unit 101. Duplicate A class code corresponding to this class tap is obtained by classifying the class into one of a number of classes.
[0039]
The class classifying unit 102 vector quantizes a tap vector (p1, p2) having, for example, pixel data p1 and p2 included in a class tap as components, using a class classification codebook, and class corresponding to the class tap. Ask for code. For example, as shown in FIG. 3, the class classification unit 102 includes a codebook storage unit 102a and a vector quantization unit 102b.
[0040]
The code book storage unit 102a stores a class classification code book obtained in advance. As shown in FIG. 4, the class classification code book includes a class corresponding to an SD code vector (P1, P2) as a representative vector for an SD signal representing each class for each of a plurality of classes. This is assignment information to which the class code (SD code) #c shown is assigned.
[0041]
In the present embodiment, since the tap vector (p1, p2) to be vector quantized is a two-dimensional vector, the SD code vector (P1, P2) is also a two-dimensional vector. A method for generating the class classification code book will be described later.
[0042]
Based on the class classification code book stored in the code book storage unit 102a, the vector quantization unit 102b performs the tap vector (p1, p2) described above. )When Multiple classes of SD code vectors (P1, P2) Between each The distance is obtained, and the class code #c assigned to the SD code vector (P1, P2) that minimizes the distance is output.
[0043]
Returning to FIG. 1, the image signal processing apparatus 100 includes a feature amount acquisition unit 103. The feature quantity acquisition unit 103 acquires a feature quantity related to the class tap based on the class tap extracted by the tap extraction unit 101. This feature amount is, for example, the dynamic range DR of the pixel data in the class tap. The dynamic range DR is a difference between the maximum value and the minimum value of the pixel data in the class tap. In addition to the dynamic range DR described above, the feature amount is a dispersion value of pixel data in a class tap, a standard deviation, a waveform pattern, difference data between a tap vector and a code vector corresponding to the class code, or the like. Also good.
[0044]
In addition, the image signal processing apparatus 100 includes a data generation unit 104. The data generation unit 104 includes the class code obtained by the class classification unit 102 and the feature amount acquired by the feature amount acquisition unit 103. Using The HD signal corresponding to the class tap extracted by the tap extraction unit 101 is configured. Double In the present embodiment, several pieces of pixel data q1 to q4 (see FIG. 2) adjacent in the horizontal direction are generated.
[0045]
For example, the data generation unit 104 includes a class code and Consisting of dynamic range Feature value Is assigned a code vector as a representative vector for the HD signal. This class code is used using the data generation code book. The An inverse code quantization is performed to obtain an HD code vector (q1, q2, q3, q4) having four pixel data q1 to q4 constituting an HD signal as components. For example, as shown in FIG. 3, the data generation unit 104 includes a codebook storage unit 104a and a vector inverse quantization unit 104b.
[0046]
The code book storage unit 104a stores a data generation code book obtained in advance. As shown in FIG. 5, this data generation code book is allocation information in which HD code vectors (Q1, Q2, Q3, Q4) as representative vectors for HD signals are assigned to combinations of class codes and feature amounts. is there. In FIG. 5, HD code #C is a class code corresponding to an HD code vector (Q1, Q2, Q3, Q4).
[0047]
Note that the data generation code book shown in FIG. 5 is an example in which the feature amount is the dynamic range DR. In this example, the feature amount is divided into two or three for each SD code #c. However, the number of division is not limited to this. The number of divisions is determined as the number of features that are set at the finest and are combined when the corresponding HD code vectors (Q1, Q2, Q3, Q4) are the same, and finally cannot be combined. It can also be defined.
[0048]
In the present embodiment, since the HD code vector (q1, q2, q3, q4) obtained by vector inverse quantization is a four-dimensional vector, the HD code vector (Q1, Q2, Q3, Q4) is also a four-dimensional vector. is there. A method for generating the data generation code book will be described later.
[0049]
The vector quantization unit 104b is a code book for data generation stored in the code book storage unit 104a. Using The HD code vector (Q1, Q2, Q3, Q4) assigned to the combination of the class code obtained by the class classification unit 102 and the feature quantity obtained by the feature quantity obtaining unit 103 is obtained, and this is obtained as an HD code vector. Output as (q1, q2, q3, q4).
[0050]
Next, the operation of the image signal processing apparatus 100 of FIG. 1 will be described.
The SD signal is supplied to the tap extraction unit 101. In the tap extraction unit 101, the SD signal is divided into blocks composed of two pieces of pixel data adjacent in the horizontal direction, and pixel data p1 and p2 (see FIG. 2) included in this block are used as class taps for each block. Extracted.
[0051]
The class tap extracted by the class tap extraction unit 101 is supplied to the class classification unit 102 and the feature amount acquisition unit 103. In the class classification unit 102, the cluster Duplicate A class code corresponding to this class tap is obtained by classifying into any one of a number of classes. In this case, for example, the tap vectors (p1, p2) having the pixel data p1, p2 included in the class tap as components are subjected to vector quantization using the class classification code book, and a class code corresponding to the class tap is obtained. It is done.
[0052]
In addition, the feature amount acquisition unit 103 uses a class tap. Find the difference between the maximum and minimum pixel data in The feature amount related to the class tap, for example, the dynamic range DR of the pixel data in the class tap is acquired.
[0053]
The class code obtained by the class classification unit 102 and the feature quantity acquired by the feature quantity acquisition unit 103 are supplied to the data generation unit 104. In this data generation unit 104, a class code and a feature amount A code book for data generation in which a code vector as a representative vector for an HD signal is assigned to the combination of Four pixel data q1 to q4 (see FIG. 2) constituting the HD signal corresponding to the class tap extracted by the tap extraction unit 101 are generated.
[0054]
This place This Class code uses the data generation codebook Is Thus, an HD code vector (q1, q2, q3, q4) having four pixel data q1 to q4 constituting the HD signal as components is obtained.
[0055]
As described above, in the image signal processing apparatus 100 shown in FIG. 1, when the SD signal is converted into the HD signal, for each block including the two pieces of pixel data p1 and p2 obtained by dividing the SD signal, Class code of class tap consisting of pixel data included in the block and Consisting of dynamic range Feature value Using , Which generates four pixel data q1 to q4 constituting an HD signal, can perform an appropriate conversion process as compared with a case based only on a class code, and can convert an SD signal into an HD signal satisfactorily. . The image signal processing apparatus 100 can be applied to an apparatus that outputs an image signal, such as a television receiver or an image signal reproduction apparatus.
[0056]
Next, the generation of the class classification code book stored in the code book storage unit 102a and the data generation code book stored in the code book storage unit 104a will be described.
[0057]
FIG. 6 shows a configuration of a code book generating apparatus 200 that generates a class classification code book and a data generation code book.
The code book generating apparatus 200 has a learning database 201. The learning database 201 stores a learning image signal as a teacher signal corresponding to the HD signal.
[0058]
In addition, the code book generating device 200 includes a student data generating unit 202 as a first tap extracting unit. The student data generation unit 202 thins out the number of pixel data of the learning image signal (teacher signal) using a thinning filter (low-pass filter) so that the number of pixel data is halved in the horizontal direction, and obtains a student signal corresponding to the SD signal. The student signal is divided into a first block composed of two pieces of pixel data p1 and p2 (see FIG. 2) adjacent in the horizontal direction, and the pixel data included in the first block is divided into the first block for each block. Extracted as one tap (student data).
[0059]
In addition, the code book generation device 200 includes a teacher data generation unit 203 as a second tap extraction unit. The teacher data generation unit 203 includes four pieces of pixel data q1 adjacent in the horizontal direction corresponding to the first block obtained by dividing the learning image signal (teacher signal) by the student data generation unit 202 described above. Is divided into second blocks each consisting of q4 (see FIG. 2), and pixel data included in the second block is extracted as a second tap (teacher data) for each block.
[0060]
Here, one first tap (student data) extracted by the student data generation unit 202 and one second tap (teacher data) extracted by the teacher data generation unit 203 corresponding thereto. A learning pair is constructed. In the code book generation device 200, a class classification code book and a data generation code book are generated using a plurality of learning pairs.
[0061]
The student data generation unit 202 obtains and uses a student signal corresponding to the SD signal by thinning out the number of pixel data of the learning image signal (teacher signal) so as to be ½ in the horizontal direction. A learning image signal as a student signal corresponding to the SD signal may also be stored in the learning database 201, and the learning image signal (student signal) may be used.
[0062]
In addition, the code book generation apparatus 200 includes a class classification unit 204. The class classification unit 204 is a first tag extracted by the student data generation unit 202. Duplicate A class code corresponding to the first tap is obtained by classifying the class into any one of the classes.
[0063]
The class classification unit 204 vector-quantizes a tap vector (p1, p2) having, for example, pixel data p1 and p2 included in the first tap as components, using a class classification codebook, and outputs the first tap. The class code corresponding to is obtained. The class classification unit 204 includes a class classification code book generation unit 204a and a vector quantization unit 204b.
[0064]
The first tap extracted by the student data generation unit 202 is supplied to the class classification code book generation unit 204a. The code book generation unit 204a uses a tap vector (p1, p2) having two pixel data p1 and p2 included in the first tap as components, for example, according to an LBG (Linde Buzo Gray) algorithm. Generate a codebook for use.
[0065]
The class classification code book is stored in a storage unit 204c built in the code book generation unit 204a, and is read from the storage unit 204c and used as necessary. This class classification code book is used by being stored in the code book storage unit 102a of FIG. 3 described above. As shown in FIG. 4, an SD representing each class for each of a plurality of classes. This is assignment information in which a class code (SD code) #c indicating a corresponding class is assigned to an SD code vector (P1, P2) as a representative vector for a signal.
[0066]
The vector quantization unit 204b is configured in the same manner as the vector quantization unit 102b of FIG. 3 described above, and the class classification code book stored in the storage unit 204c of the code book generation unit 204a. Using, Tap vector (p1, p2) )When, Multiple classes of SD code vectors (P1, P2) Between each And the class code #c assigned to the SD code vector (P1, P2) that minimizes the distance is output.
[0067]
In addition, the code book generating apparatus 200 includes a feature amount acquisition unit 205. The feature amount acquisition unit 205 is configured in the same manner as the feature amount acquisition unit 103 of FIGS. 1 and 3 described above, and the first tap extracted by the student data generation unit 202 is used. To The feature amount is acquired. This feature amount is, for example, the dynamic range DR of the pixel data in the class tap.
[0068]
In addition, the code book generating apparatus 200 includes a code book generating unit 206. The code book generator 206 is supplied with the second tap extracted by the teacher data generator 203. The code book generation unit 206 uses the tap vector (q1, q2, q3, q4) whose components are the four pixel data q1 to q4 included in the second tap, for example, in accordance with the LBG algorithm. Generate such a codebook.
[0069]
As shown in FIG. 7, this code book includes a class corresponding to an HD code vector (Q1, Q2, Q3, Q4) as a representative vector for an HD signal representing each class for each of a plurality of classes. This is assignment information to which a class code (HD code) #C indicating the assignment is assigned. This code book corresponds to a class classification code book for HD signals. The code book is stored in a storage unit 206a built in the code book generation unit 206, and is read from the storage unit 206a and used as necessary.
[0070]
In addition, the code book generating apparatus 200 includes an optimum link detecting unit 207 as a code book generating unit. The optimum link detection unit 207 includes a class code obtained by the class classification unit 204, a feature amount obtained by the feature amount acquisition unit 205, a second tap extracted by the teacher data generation unit 203, and a code book generation unit. Code book generated in 206 Using, Class code and Consisting of dynamic range A code book for data generation is assigned by assigning one HD code vector (Q1, Q2, Q3, Q4) selected from a plurality of HD code vectors (Q1, Q2, Q3, Q4) to each combination of feature amounts. Generate.
[0071]
The data generation code book generated by the optimum link detection unit 207 is stored in the storage unit 207a built in the optimum link detection unit 207, and is read from the storage unit 207a and used as necessary. . This data generation code book is used by being stored in the code book storage unit 104a of FIG. 3 described above, and, as shown in FIG. This is assignment information in which HD code vectors (Q1, Q2, Q3, Q4) as vectors are assigned.
[0072]
In this case, the optimum link detection unit 207 specifically performs the following processing, for example. That is, for each combination of the class code and the feature amount, each of the plurality of second taps corresponding to the combination, that is, the plurality of second taps corresponding to the plurality of first taps from which the class code and the feature amount of the combination are obtained. Error between the pixel data q1 to q4 included in the 2 taps and the components Q1 to Q4 of the HD code vector (Q1, Q2, Q3, Q4) (for example, the difference obtained by squaring the difference for each pixel data) The sum of squares) is obtained, and the sum of the obtained plurality of errors is obtained to obtain an error sum by a plurality of HD code vectors, and HD code vectors (Q1, Q2, Q3, which minimize the error sum) are obtained. Q4) is assigned to the combination.
[0073]
Here, assuming that there are M first taps in a certain combination, the pixel data included in the second tap corresponding to the m-th first tap is represented as q1 (m) to q4 (m). I will do it. In this case, the error sum corresponding to a single HD code vector (Q1, Q2, Q3, Q4) is expressed by, for example, equation (1). Here, Σ represents the summation with m changed from 1 to M.
Error sum = Σ {(q1 (m) −Q1) ² + (Q2 (m) -Q2) ² + (Q3 (m) -Q3) ² + (Q4 (m) -Q4) ² } ... (1)
[0074]
Next, the operation of the code book generating apparatus 200 shown in FIG. 6 will be described.
A learning image signal as a teacher signal corresponding to the HD signal stored in the learning database 201 is supplied to the student data generation unit 202. In the student data generation unit 202, the first tap is extracted based on the learning image signal (teacher signal).
[0075]
That is, thinning is performed using a thinning filter so that the number of pixel data of the learning image signal is halved in the horizontal direction, a student signal corresponding to the SD signal is obtained, and the student signal is adjacent in the horizontal direction. Are divided into a first block composed of two pieces of pixel data p1 and p2 (see FIG. 2), and pixel data included in the first block is extracted as a first tap (student data) for each block. The
[0076]
Further, a learning image signal as a teacher signal corresponding to the HD signal stored in the learning database 201 is supplied to the teacher data generation unit 203. In the teacher data generation unit 203, the second tap is extracted based on the learning image signal (teacher signal). That is, four pieces of pixel data q1 to q4 adjacent to the horizontal direction corresponding to the first block obtained by dividing the learning image signal (teacher signal) by the student data generation unit 202 described above (see FIG. 2). ), And pixel data included in the second block is extracted as a second tap (teacher data) for each block.
[0077]
The first tap extracted by the student data generation unit 202 is supplied to the class classification code book generation unit 204 a constituting the class classification unit 204. The code book generation unit 204a uses a tap vector (p1, p2) whose components are two pieces of pixel data p1, p2 included in the first tap, for example, according to the LBG algorithm (see FIG. 4) is generated.
[0078]
This class classification code book is stored and used in the code book storage unit 102a of FIG. 3 described above. The class classification code book is stored in a storage unit 204c built in the code book generation unit 204a, and is read from the storage unit 204c and used as necessary.
[0079]
The second tap extracted by the teacher data generation unit 203 is supplied to the code book generation unit 206. The code book generation unit 206 uses a tap vector (q1, q2, q3, q4) having four pixel data q1 to q4 included in the second tap as components, for example, according to the LBG algorithm, Is generated). The code book is stored in a storage unit 206a built in the code book generation unit 206, and is read from the storage unit 206a and used as necessary.
[0080]
Further, the first tap extracted by the student data generation unit 202 is supplied to the vector quantization unit 204 b constituting the class classification unit 204. In the class classification unit 204, the first tap Is The class code corresponding to the first tap is obtained. That is, in the vector quantization unit 204b, the tap classification (p1, p2) having the pixel data p1 and p2 included in the first tap as components is stored in the storage unit 204c of the codebook generation unit 204a. The class code corresponding to the first tap is obtained by vector quantization using the code book.
[0081]
Further, the first tap extracted by the student data generation unit 202 is supplied to the feature amount acquisition unit 205. In the feature amount acquisition unit 205, the first touch In charge For example, the dynamic range DR of the pixel data in the first tap.
[0082]
The class code obtained by the vector quantization unit 204b and the feature quantity acquired by the feature quantity acquisition unit 205 are supplied to the optimum link detection unit 207. Further, the optimum link detection unit 207 is also supplied with the second tap extracted by the teacher data generation unit 203 and the code book stored in the storage unit 206a of the code book generation unit 206.
[0083]
In the optimum link detection unit 207, the class code, the feature amount, the second tap, and the code book Using , Class code and Consisting of dynamic range One HD code vector (Q1, Q2, Q3, Q4) selected from a plurality of HD code vectors (Q1, Q2, Q3, Q4) is assigned to each combination of feature amounts, and a data generation codebook (See FIG. 5) is generated.
[0084]
In this case, for each combination of the class code and the feature amount, pixel data q1 to q4 included in the second tap for each of the plurality of second taps corresponding to the plurality of first taps corresponding to the combination, and An error is obtained between the components Q1 to Q4 of the HD code vector (Q1, Q2, Q3, Q4), and the obtained plurality of errors are added to obtain an error sum (see equation (1)). This is repeated for a plurality of HD code vectors, and HD code vectors (Q1, Q2, Q3, Q4) that minimize the error sum are assigned to the combination.
[0085]
This data generation code book is used by being stored in the code book storage unit 104a of FIG. 3 described above. The data generation code book is stored in a storage unit 207a built in the code book generation unit 207, and is read from the storage unit 207a and used as necessary.
[0086]
As described above, according to the code book generating apparatus 200 shown in FIG. 6, the class classification code book and the data generating code book used in the image signal processing apparatus 100 of FIG. 1 can be generated satisfactorily.
[0087]
In the code book generating device 200 shown in FIG. 6, the class classification code book generating unit 204a and the code book generating unit 206 of the class classifying unit 204 generate code books according to, for example, the LBG algorithm. The code book may be generated by a method other than the method used for generating the code book used for vector quantization.
[0088]
That is, the class classification code book can be generated, for example, as shown in FIG.
[0089]
Here, in FIG. 8, a two-dimensional space (SD image space) representing the two pieces of pixel data p1 and p2 included in the first tap is divided into a lattice shape, thereby obtaining a rectangular region r (i , j) is configured. In FIG. 8, the pixel values of the pixel data p1 and p2 are plotted on the horizontal axis and the vertical axis, respectively.
[0090]
In the present embodiment, since the pixel data p1 and p2 are 8-bit data as described above, the horizontal axis and the vertical axis in the SD image space can take values from 0 to 255 (= 2). ⁸ -1). In FIG. 8, r (i, j) represents the i-th region from the left and the j-th region from the bottom in the SD image space, and each region r (i, j) represents a unique class. Class code #c is assigned.
[0091]
The class classification code book generation unit 204a obtains, for each region r (i, j), the first tap included in the region r (i, j), and further, for example, the first tap of the region t (i, j). Find the center of gravity. That is, there are K first taps included in the region r (i, j), and the k-th first tap is (p _{1, r (i, j)} (k), p _{2, r (i, j)} (k)), the class classification code book generation unit 204a is represented by (Σp _{1, r (i, j)} (k) / K, Σp _{, r (i, j)} The center of gravity of the first tap of the class code #c assigned to the region r (i, j) represented by (k) / K) is obtained. Here, Σ represents a summation with k changed from 1 to K.
[0092]
Then, the class classification code book generation unit 204a uses the SD code vector (P1, P2) for the SD signal to represent the center of gravity of the first tap of the class code #c and the class represented by the class code #c ) To correspond to the class code #c. As a result, a class classification code book as shown in FIG. 4 is generated.
[0093]
Although detailed description is omitted, the code book (see FIG. 7) generated by the code book generation unit 206 can also be generated in the same manner as the class classification code book described above.
[0094]
In that case, the region r (i, j, s, t) is configured by dividing the four-dimensional space (HD image space) representing the four pieces of pixel data q1 to q4 included in the second tap. Is done.
[0095]
In this case, r (i, j, s, t) is i-th on the axis of the pixel value of the image data q1 and j-th on the axis of the pixel value of the image data q2 in the HD image space. The pixel value axis represents the s-th region and the pixel value axis of the image data q4 represents the t-th region, and each region r (i, j, s, t) has a unique class code #C representing a class. Assigned.
[0096]
For each region r (i, j, s, t), the codebook generation unit 206 obtains second taps included in the region r (i, j, s, t), and those second taps Find the center of gravity. Then, the code book generating unit 206 determines the center of gravity of the second tap of the class code #C as the HD code vector (Q1, Q2, Q3) for the HD signal representing the class represented by the class code #C. Q4) is associated with class code #C. Thereby, a code book as shown in FIG. 7 is generated.
[0097]
Note that the above-described processing in the image signal processing apparatus 100 of FIG. 1 can also be performed by software using, for example, an image signal processing apparatus (computer) 500 as shown in FIG.
[0098]
First, the image signal processing apparatus 500 shown in FIG. 9 will be described. The image signal processing apparatus 500 includes a CPU 501 that controls the operation of the entire apparatus, a ROM (Read Only Memory) 502 that stores a control program for the CPU 501 and a code book, and a RAM (Random) that constitutes a work area of the CPU 501. Access Memory) 503. These CPU 501, ROM 502 and RAM 503 are each connected to a bus 504.
[0099]
The image signal processing apparatus 500 includes a hard disk drive (HDD) 505 as an external storage device, a flexible disk, a CD-ROM (Compact Disc Read only Memory), an MO (Magneto Optical) disk, a DVD (Digital Versatile Disc), And a drive 506 for handling a removable recording medium such as a magnetic disk or a semiconductor memory. These drives 505 and 506 are connected to a bus 504, respectively.
[0100]
Further, the image signal processing apparatus 500 includes a communication unit 508 that is connected to a communication network 507 such as the Internet by wire or wirelessly. The communication unit 508 is connected to the bus 504 via the interface 509.
[0101]
Further, the image signal processing apparatus 500 includes a user interface unit. The user interface unit includes a remote control signal receiving circuit 511 that receives a remote control signal RM from the remote control transmitter 510, and a display 513 that includes a CRT (Cathode-Ray Tube), an LCD (liquid Crystal Display), or the like. . The receiving circuit 511 is connected to the bus 504 via the interface 512, and similarly the display 513 is connected to the bus 504 via the interface 514.
[0102]
The image signal processing apparatus 500 also has an input terminal 515 for inputting an SD signal and an output terminal 517 for outputting an HD signal. Input terminal 515 is connected to bus 504 via interface 516, and similarly output terminal 517 is connected to bus 504 via interface 518.
[0103]
Here, instead of storing the control program or code book in the ROM 502 in advance as described above, for example, it is downloaded from the communication network 507 such as the Internet via the communication unit 508 and stored in the hard disk drive 505 or RAM 303. It can also be used. Also, these control programs, code books, etc. may be provided on a removable recording medium.
[0104]
Further, instead of inputting the SD signal to be processed from the input terminal 515, the SD signal may be supplied by a removable recording medium or may be downloaded from the communication network 507 such as the Internet via the communication unit 508. Further, instead of outputting the processed HD signal to the output terminal 517 or in parallel therewith, it is supplied to the display 513 for image display, further stored in the hard disk drive 505, or via the communication unit 508 via the Internet. It may be transmitted to the communication network 507.
[0105]
A processing procedure for obtaining the HD signal from the SD signal in the image signal processing apparatus 500 shown in FIG. 9 will be described with reference to the flowchart of FIG.
[0106]
First, in step ST11, processing is started, and in step ST12, for example, an SD signal for one frame or one field is input into the apparatus from the input terminal 515. The SD signal input in this way is temporarily stored in the RAM 503.
[0107]
Then, in step ST13, it is determined whether or not processing of all frames or all fields of the SD signal has been completed. When the process is finished, the process ends in step ST14. On the other hand, when the process is not finished, the process proceeds to step ST15.
[0108]
In step ST15, the SD signal is divided into blocks composed of two pixel data adjacent in the horizontal direction, and pixel data p1 and p2 (see FIG. 2) included in this block are extracted as class taps for each block. The
[0109]
In step ST16, the cluster Duplicate Class code is obtained by classifying into any one of a number of classes. In this case, the tap vector (p1, p2) having the pixel data p1, p2 included in the class tap as components is subjected to vector quantization using the class classification codebook (see FIG. 4) stored in the ROM 502, The class code corresponding to the class tap is obtained.
[0110]
Next, in step ST17, the cluster Related to The feature amount, for example, the dynamic range DR of the pixel data p1 and p2 in the class tap is acquired.
[0111]
Next, in step ST18, the class code obtained in step ST16 and obtained in step ST17. Consisting of dynamic range Feature value Using Then, four pixel data q1 to q4 (see FIG. 2) constituting the HD signal corresponding to the class tap extracted in step ST15 are generated. This place This The class code is vector-dequantized using a data generation code book (see FIG. 5) stored in the ROM 502, and an HD code vector having four pixel data q1 to q4 constituting an HD signal as components ( q1, q2, q3, q4) are obtained.
[0112]
Next, in step ST19, it is determined whether or not the processing for obtaining the pixel data of the HD signal has been completed in the entire area of the SD signal pixel data for one frame or one field input in step ST12. If completed, the process returns to step ST12, and the process proceeds to SD signal input processing for the next one frame or one field. On the other hand, when the process has not been completed, the process returns to step ST15 and proceeds to the process for the next class tap.
[0113]
In this way, by performing processing according to the flowchart shown in FIG. 10, the pixel data of the input SD signal can be processed to obtain the pixel data of the HD signal. The HD signal obtained by such processing is output to the output terminal 517, supplied to the display 513 to display an image, and further supplied to the hard disk drive 505 to be recorded.
[0114]
Further, although illustration of the processing device is omitted, the processing in the code book generation device 200 shown in FIG. 6 can also be performed by software.
[0115]
A processing procedure for generating a class classification code book and a data generation code book will be described with reference to the flowchart of FIG.
First, processing is started in step ST21, and in step ST22, a teacher signal corresponding to the HD signal is input for one frame or one field. In step ST23, a student signal corresponding to the SD signal is generated from the teacher signal input in step ST22. In this case, the student signal corresponding to the SD signal is generated by thinning out the number of pixel data of the teacher signal corresponding to the HD signal so as to be ½ in the horizontal direction.
[0116]
Next, in step ST24, the student information generated in step ST23 is displayed. No. 1 tap (student data) As Extract. In this case, the student signal is divided into first blocks composed of two pieces of pixel data p1 and p2 (see FIG. 2) adjacent in the horizontal direction, and the pixel data included in the first block is divided into the first block for each block. Extract as 1 tap.
[0117]
Next, in step ST25, the teacher signal input in step ST22. No. 2 taps (teacher data) As Extract. In this case, the teacher signal is assigned to a second block composed of four pieces of pixel data q1 to q4 (see FIG. 2) adjacent in the horizontal direction, corresponding to the first block obtained by dividing in step ST24 described above. The pixel data is divided and the pixel data included in the second block is extracted as the second tap for each block.
[0118]
By the processing of step ST24 and step ST25, a plurality of pairs of learning data is generated from the teacher signal and student signal for one frame or one field. Incidentally, one learning data is composed of one first tap (student data) and one second tap (teacher data).
[0119]
Next, in step ST26, it is determined whether or not processing of all the frames or fields of the teacher signal has been completed. If not completed, the process returns to step ST22, the teacher signal for the next one frame or one field is input, and the same processing as described above is repeated. On the other hand, when it is finished, the process proceeds to step ST27.
[0120]
In step ST27, a class classification code book is generated. In this case, based on a plurality of first taps (student data) constituting a plurality of pairs of learning data generated by the processing of step ST22 to step ST26, two pieces of pixel data p1, included in the first tap Using a tap vector (p1, p2) having p2 as a component, a class classification codebook is generated, for example, according to the LBG algorithm.
[0121]
As shown in FIG. 4, the class classification code book includes a class corresponding to an SD code vector (P1, P2) as a representative vector for an SD signal representing each class for each of a plurality of classes. This is assignment information to which the class code (SD code) #c shown is assigned.
[0122]
Next, in step ST28, a code book related to the HD signal is generated. In this case, based on a plurality of second taps (teacher data) constituting a plurality of pairs of learning data generated by the processing of step ST22 to step ST26, four pieces of pixel data q1 to q1 included in the second tap are included. Using a tap vector (q1, q2, q3, q4) having q4 as a component, a code book is generated, for example, according to the LBG algorithm.
[0123]
As shown in FIG. 7, this class code is a class corresponding to an HD code vector (Q1, Q2, Q3, Q4) as a representative vector for an HD signal that represents each class for each of a plurality of classes. Is allocation information to which a class code (HD code) #C indicating the
[0124]
Next, in step ST29, based on a plurality of first taps (student data) constituting a plurality of pairs of learning data generated by the processing in steps ST22 to ST26, each first tap is selected from a plurality of classes. And class codes corresponding to the first taps are obtained.
[0125]
In this case, the tap vector (p1, p2) having the pixel data p1, p2 included in the first tap as components is vector-quantized using the class classification codebook generated in step ST27, and the first The class code #c corresponding to the tap is obtained.
[0126]
Next, in step ST30, based on a plurality of first taps (student data) constituting a plurality of pairs of learning data generated by the processing in steps ST22 to ST26, for example, feature amounts related to the first taps, for example, The dynamic range of the pixel data in the first tap is acquired.
[0127]
Next, in step ST31, a plurality of pairs of learning data generated by the processing in steps ST22 to ST26, a code book (see FIG. 7) generated in step ST28, and a plurality of pairs generated in steps ST29 and ST30. Class code #c and feature quantity #a corresponding to each first tap (student data) constituting the learning data Using A data creation code book is generated, and the process ends in step ST32.
[0128]
Here, the feature amount #a indicates the number of each segmentation range in the finest segmentation of the feature amount acquired in step ST30. For example, when the feature amount is the dynamic range DR, the dynamic range DR is in the range of 0 to 255, and the finest division is performed for each “1”, there are 256 division ranges, and thus the feature amount # a changes from 0 to 255.
[0129]
As shown in FIG. 5, the data generation codebook includes a plurality of codebooks registered in the codebook (see FIG. 7) generated in step ST28 for each combination of the class code #c and the feature quantity #a. Allocation information to which one HD code vector (Q1, Q2, Q3, Q4) selected from the HD code vectors (Q1, Q2, Q3, Q4) is allocated.
[0130]
In the process of step ST31, for each combination of class code #c and feature quantity #a, each of a plurality of second taps corresponding to a plurality of first taps corresponding to the combination is included in the second tap. An error is obtained between the pixel data q1 to q4 to be generated and the components Q1 to Q4 of the HD code vector (Q1, Q2, Q3, Q4), and a plurality of the obtained errors are added to obtain an error sum (equation (1) Is obtained for a plurality of HD code vectors, and HD code vectors (Q1, Q2, Q3, Q4) that minimize the error sum are assigned to the combinations.
[0131]
That is, first, in step ST41, the class code (SD code) #c is set to 0. In step ST42, the feature quantity #a is set to 0. In step ST43, the class code (HD code) #C is set to 0. Set to.
[0132]
In step ST44, an error sum corresponding to the set class code #C in the combination of the set class code #c and the feature quantity #a is calculated. In this case, pixel data q <b> 1 to q <b> 1 included in the second tap for each of the plurality of second taps corresponding to the plurality of first taps corresponding to the set combination of the class code #c and the feature amount #a. An error is obtained between q4 and the components Q1 to Q4 of the HD code vector (Q1, Q2, Q3, Q4) corresponding to the set class code #C, and an error is obtained by adding the obtained plural errors. Find the sum.
[0133]
Next, in step ST45, it is determined whether or not the processing for calculating the error sum corresponding to all the class codes (HD codes) #C in the set combination of the class code #c and the feature quantity #a is completed. To do. If the process has not been completed, the class code #C is incremented in step ST46, and then the process returns to step ST44, and the next class code #C in the set combination of the class code #c and the feature quantity #a. The process proceeds to obtain an error sum corresponding to. On the other hand, when the process ends, the process proceeds to step ST47.
[0134]
In step ST47, is the processing for calculating the error sum corresponding to all the class codes (HD codes) #C in the combination of the set class code (SD code) #c and all the feature values #a completed? Determine whether or not. If the process has not been completed, the feature quantity #a is incremented in step ST48, and then the process returns to step ST43, in the combination of the set class code (SD code) #c and the next feature quantity #a. Then, the process proceeds to a process of obtaining error sums corresponding to all class codes (HD codes) #C. On the other hand, when the process ends, the process proceeds to step ST49.
[0135]
In step ST49, is the processing for calculating the sum of errors corresponding to all class codes (HD codes) #C in all class codes (SD codes) #c in combination with all feature values #a completed? Determine whether or not. If the process has not been completed, the class code #c is incremented in step ST50, and then the process returns to step ST42, and the next class code (SD code) #c is used in combination with all the feature values #a. Then, the process proceeds to a process of obtaining error sums corresponding to all class codes (HD codes) #C. On the other hand, when the process ends, the process proceeds to step ST51.
[0136]
When it is determined in step ST49 that the process has been completed, an error sum corresponding to all class codes (HD codes) #C in each combination of the class code #c and the feature quantity #a is obtained. In step ST51, for each combination of class code #c and feature quantity #a, an HD code vector (Q1, Q2, Q3, Q4) corresponding to class code #C that minimizes the error sum is obtained, and this combination is obtained. And a data generation code book (see FIG. 5) is generated.
[0137]
In this manner, by performing processing according to the flowchart shown in FIG. 11, the class classification code book and the data generation code book can be generated by the same method as the code book generation apparatus 200 shown in FIG.
[0138]
In the image signal processing apparatus 100 shown in FIG. 1, the SD signal is divided into blocks each composed of two pixel data adjacent in the horizontal direction, and the pixel data p1, p2 ( 2) is extracted as a class tap, and for each class tap, four pieces of pixel data q1 to q4 (see FIG. 2) adjacent to the horizontal direction constituting the HD signal are generated. The number and arrangement of the pixel data of the HD signal generated corresponding to this class tap is not limited to this. That is, the number of pixel data included in the class tap is not limited to two, and may not be adjacent in the horizontal direction. The pixel data of the HD signal generated corresponding to the class tap is not limited to four, and may not be adjacent in the horizontal direction.
[0139]
Further, in the image signal processing apparatus 100 shown in FIG. 1, the tap vectors (p1, p2) having the pixel data p1, p2 included in the class tap as components are vector quantized using a class classification codebook, Although the method for obtaining the class code corresponding to the class tap is shown, the method for obtaining the class code corresponding to the class tap is not limited to this. That is, this class classification can be obtained by performing the above-described ADRC processing on the pixel data itself constituting the class tap or data obtained by processing it, for example, the difference value between adjacent pixel data. It is. In that case, the class classification unit 204 in the code book generating apparatus 200 shown in FIG.
[0140]
In the image signal processing apparatus 100 shown in FIG. The Lascord De Using the data generation code book, vector dequantization is performed to obtain an HD code vector (q1, q2, q3, q4) having four pixel data q1 to q4 constituting an HD signal as components. But the class code and Consisting of dynamic range The method of generating pixel data constituting the HD signal corresponding to the feature amount is not limited to this. For example, it is possible to calculate pixel data constituting the HD signal from the class code and the feature quantity with a predetermined relational expression. In this case, the coefficient of the predetermined relational expression can be generated in advance by learning.
[0141]
In the image signal processing apparatus 100 shown in FIG. 1, the SD signal is converted into an HD signal in which the number of pixels is doubled in the horizontal direction. However, the direction of increasing the number of pixels is limited to the horizontal direction. The vertical direction and further the time direction (frame direction) are also conceivable. Conversely, the present invention can be similarly applied when obtaining an SD signal with a reduced number of pixels from an HD signal. In other words, the present invention is generally applicable to the case where the first image signal is converted into the second image signal having the same or different number of pixels as the first image signal.
[0142]
In the above-described embodiment, the information signal composed of a plurality of information data is an image signal composed of a plurality of pixel data. However, in the present invention, the information signal is other, for example, an audio signal. The same applies to things. In the case of an audio signal, it consists of a plurality of sample data.
[0143]
【The invention's effect】
In this invention Information signal processing apparatus, information signal processing method and program According to the first information signal comprising a plurality of information data, Many sample pixels When converting to a second information signal consisting of a plurality of information data, the present invention Since the code book for data generation created by the code book creation device, the code book creation method and the program is used, The first information signal High resolution Converts well to the second information signal Become .
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of an image signal processing apparatus.
FIG. 2 is a diagram illustrating a pixel position relationship between an SD signal and an HD signal.
FIG. 3 is a block diagram illustrating a configuration of a class classification unit and a data generation unit.
FIG. 4 is a diagram illustrating an example of a class classification code book.
FIG. 5 is a diagram illustrating an example of a data generation code book.
FIG. 6 is a block diagram illustrating a configuration of a code book generating device.
FIG. 7 is a diagram illustrating an example of a code book related to an HD signal.
FIG. 8 is a diagram for explaining another example of a code book generation method.
FIG. 9 is a block diagram illustrating a configuration of an image signal processing device to be realized by software.
FIG. 10 is a flowchart showing image signal processing.
FIG. 11 is a flowchart illustrating a code book generation process.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 100 ... Image signal processing apparatus, 101 ... Tap extraction part, 102 ... Class classification part, 102a ... Codebook storage part, 102b ... Vector quantization part, 103 ... Acquisition of feature-value 104, data generation unit, 104a, codebook storage unit, 104b, vector inverse quantization unit, 200, codebook generation device, 201, learning database, 202,. Student data generation unit, 203 ... teacher data generation unit, 204 ... class classification unit, 204a ... class classification codebook generation unit, 204b ... vector quantization unit, 204c ... storage unit, 205: Feature amount acquisition unit, 206: Code book generation unit, 206a: Storage unit, 207: Optimal link detection unit, 207a: Storage unit, 500: Image Signal processing device

Claims (13)

An information signal processing device for converting a first information signal composed of a plurality of information data into a second information signal composed of a plurality of information data having a larger number of sample pixels than the first information signal,
Class tap extracting means for dividing the first information signal into blocks each including a plurality of information data, and for each block, extracting information data included in the block as a class tap;
Class classifying means for classifying the class tap extracted by the class tap extracting means into any one of a plurality of classes based on the signal value of each class tap and obtaining a class code representing the feature of the block When,
A feature quantity acquisition means for obtaining a dynamic quantity that is a difference between a maximum value and a minimum value of pixel data in the class tap extracted by the class tap extraction means to obtain a feature quantity related to the class tap;
The representative vector assigned to the combination of the feature code composed of the class code obtained by the class classification means and the dynamic range obtained by the feature quantity obtaining means is obtained from the data generation codebook, and corresponds to the class tap. , e Bei and data generating means for generating a plurality of information data that constitutes the second information signal,
The classification means is
The tap vector to the extracted information data contained in the class tap components in the class tap extraction means, using the class classification codebook, vector-quantized, the Rukoto obtain a class code corresponding to the class tap A characteristic information signal processing apparatus. The classification means is
Storage means for storing a class classification code book in which a class code indicating a corresponding class is assigned to a representative vector for the first information signal representing each class for each of the plurality of classes;
Vector quantization for obtaining a distance between the tap vector and each of representative vectors of a plurality of classes using the class classification codebook and outputting a class code assigned to the representative vector that minimizes the distance The information signal processing apparatus according to claim 1 , further comprising: means. The data generation means includes
A data generation code in which a code vector as a representative vector for the second information signal is assigned to a combination of a feature code composed of a class code obtained by the class classification means and a dynamic range obtained by the feature quantity acquisition means The information signal according to claim 1, wherein the class code is vector-dequantized using a book to obtain a code vector having a plurality of pieces of information data constituting the second information signal as components. Processing equipment. The data generation means includes
Storage means for storing the data generation codebook;
Using the data generation code book, obtain the representative vector assigned to the combination of the feature code consisting of the class code obtained by the class classification means and the dynamic range obtained by the feature quantity acquisition means, The information signal processing apparatus according to claim 3 , further comprising vector dequantization means for outputting the obtained representative vector as the code vector. The information signal processing apparatus according to claim 3 , wherein the data generation code book is obtained in advance by learning. The information signal processing apparatus according to claim 1, wherein the first and second information signals are image signals or audio signals. An information signal processing method for converting a first information signal composed of a plurality of information data into a second information signal composed of a plurality of information data having a larger number of sample pixels than the first information signal,
Class tap extraction step of dividing the first information signal into blocks composed of a plurality of information data, and extracting the information data included in the block as a class tap for each block;
Class classification step of classifying the class tap extracted in the class tap extraction step into any one of a plurality of classes based on the signal value of each class tap to obtain a class code representing the feature of the block When,
A feature amount obtaining step of obtaining a feature amount relating to the class tap by obtaining a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the class tap extracted in the class tap extraction step;
The representative vector assigned to the combination of the feature code consisting of the class code obtained in the class classification step and the dynamic range obtained in the feature amount acquisition step is obtained from the data generation codebook, and corresponds to the class tap. , have a data generating step of generating a plurality of information data that constitutes the second information signal,
For the above classification process,
Includes processing to quantize a tap vector whose component is information data contained in the class tap extracted in the class tap extraction step using a class classification code book to obtain a class code corresponding to the class tap. information signal processing method according to claim Rukoto Re. A program for converting a first information signal composed of a plurality of information data into a second information signal composed of a plurality of information data having a larger number of sample pixels than the first information signal,
Class tap extraction step of dividing the first information signal into blocks composed of a plurality of information data, and outputting, for each block, the information data included in the block as a class tap;
Class classification step of classifying the class tap extracted in the class tap extraction step into any one of a plurality of classes based on the signal value of each class tap to obtain a class code representing the feature of the block When,
A feature amount obtaining step of obtaining a feature amount relating to the class tap by obtaining a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the class tap extracted in the class tap extraction step;
The representative vector assigned to the combination of the feature code consisting of the class code obtained by the class classification means and the dynamic range obtained by the feature quantity acquisition means is obtained from the data generation codebook, and corresponds to the class tap. , have a data generating step of generating a plurality of information data that constitutes the second information signal,
For the above classification process,
Includes processing to quantize a tap vector whose component is the information data contained in the class tap extracted in the class tap extraction step using a class classification codebook to obtain a class code corresponding to the class tap program, characterized in that it can execute the information signal processing method Re that a computer. A codebook used when converting a first information signal composed of a plurality of information data into a second information signal composed of a plurality of information data having a larger number of sample pixels than the first information signal. Then, for each block composed of a plurality of information data obtained by dividing the first information signal, a combination of the class code of the class tap composed of information data included in the block and the feature amount is combined with the second In a code book generating device for generating a data generation code book to which a code vector as a representative vector for the information signal is assigned,
First tap extracting means for extracting, as a first tap, information data included in each first block made up of a plurality of information data obtained by dividing a student signal corresponding to the first information signal When,
The information data contained in each second block made up of a plurality of information data obtained by dividing the teacher signal corresponding to the second information signal corresponding to the first block is the second tap. Second tap extraction means for extracting as:
Classifying the first tap extracted by the first tap extracting means into one of a plurality of classes based on the signal value of each tap , and class code representing the feature of the block Classifying means to obtain;
Feature quantity acquisition for obtaining a dynamic quantity that is a difference between a maximum value and a minimum value of pixel data in the first tap extracted by the first tap extraction means, and acquiring a feature quantity related to the first tap. Means,
Storage means for storing a plurality of representative vectors whose components are a plurality of information data corresponding to the second taps extracted by the second tap extraction means for the second information signal, respectively.
Stored in the class code obtained by the class classification means, the feature quantity consisting of the dynamic range obtained by the feature quantity acquisition means, the second tap extracted by the second tap extraction means, and the storage means A codebook for generating the data generation codebook by allocating one representative vector selected from the plurality of representative vectors to each of the combination of the class code and the feature quantity based on the plurality of representative vectors. Bei example and generation means,
The classification means is
A tap vector whose component is information data included in the first tap extracted by the first class tap extraction means is vector quantized using a class classification codebook, and a class code corresponding to the class tap is obtained. codebook generation apparatus according to claim Rukoto give. The code book generation means includes
For each combination of the class code and the feature quantity, information data included in the second tap and the representative for each of the plurality of second taps corresponding to the plurality of first taps corresponding to the combination. Obtaining an error with a component in the vector, adding the obtained plurality of errors to obtain an error sum is repeated for the plurality of representative vectors, and one representative vector that minimizes the error sum is obtained. The code book generation device according to claim 9 , wherein the code book generation device is assigned to the combination. The classification means is
Storage means for storing a class classification code book in which a class code indicating a corresponding class is assigned to a representative vector for the first information signal representing each class for each of the plurality of classes;
Using the class classification code book, the distance between a vector having the information data included in the first tap as a component and the representative vectors of the plurality of classes is obtained, and the representative that minimizes the distance is obtained. The code book generation device according to claim 9 , further comprising vector quantization means for outputting a class code assigned to the vector. A codebook used when converting a first information signal composed of a plurality of information data into a second information signal composed of a plurality of information data having a larger number of sample pixels than the first information signal. Then, for each block composed of a plurality of information data obtained by dividing the first information signal, a combination of the class code of the class tap composed of the information data included in the block and the feature amount is added to the second information signal. In a code book generating method for generating a data generation code book to which a code vector as a representative vector for the information signal is assigned,
A first tap extracting step of extracting, as a first tap, information data included in each first block composed of a plurality of information data obtained by dividing a student signal corresponding to the first information signal; ,
The information data contained in each second block made up of a plurality of information data obtained by dividing the teacher signal corresponding to the second information signal corresponding to the first block is the second tap. A second tap extraction step to extract as
Classifying the first tap extracted in the first tap extraction step into one of a plurality of classes based on the signal value of each of the taps , and class code representing the feature of the block A classification process to obtain;
Feature amount acquisition for obtaining a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the first tap extracted in the first tap extraction step and acquiring a feature amount related to the first tap. Process,
The class code obtained in the class classification step, the feature amount consisting of the dynamic range obtained in the feature amount acquisition step, the second tap extracted in the second tap extraction step, and the second information signal And a combination of the class code and the feature amount based on a plurality of representative vectors each having a plurality of information data as components corresponding to the second tap extracted in the second tap extraction step. of each, by assigning one representative vectors selected from the plurality of representative vectors have a codebook generation step of generating a code book for the data generation,
For the above classification process,
A tap vector whose component is information data included in the first tap extracted by the first class tap extraction means is vector quantized using a class classification codebook, and a class code corresponding to the class tap is obtained. codebook generation method comprising Rukoto include processes to obtain. A codebook used when converting a first information signal composed of a plurality of information data into a second information signal composed of a plurality of information data having a larger number of sample pixels than the first information signal. For each block composed of a plurality of information data obtained by dividing the first information signal, a combination of the class code and the feature amount of the class tap composed of information data included in the block is combined with the second A program for generating a data generation codebook to which a code vector as a representative vector for an information signal is assigned,
A first tap extracting step of extracting, as a first tap, information data included in each first block composed of a plurality of information data obtained by dividing a student signal corresponding to the first information signal;
The information data contained in each second block made up of a plurality of information data obtained by dividing the teacher signal corresponding to the second information signal corresponding to the first block is the second tap. A second tap extraction step to extract as
Classifying the first tap extracted in the first tap extraction step into one of a plurality of classes based on the signal value of each of the taps , and class code representing the feature of the block A classification process to obtain;
Feature amount acquisition for obtaining a dynamic range that is a difference between the maximum value and the minimum value of the pixel data in the first tap extracted in the first tap extraction step and acquiring a feature amount related to the first tap. Process,
About the class code obtained in the class classification step, the feature amount consisting of the dynamic range obtained in the feature amount acquisition step, the second tap extracted in the second tap extraction step, and the second information signal Based on a plurality of representative vectors each having a plurality of pieces of information data corresponding to the second tap extracted in the second tap extraction step, respectively, by assigning a single representative vectors selected from the plurality of representative vectors have a codebook generation step of generating a code book for the data generation,
For the above classification process,
A tap vector whose component is information data included in the first tap extracted by the first class tap extraction means is vector quantized using a class classification codebook, and a class code corresponding to the class tap is obtained. a program characterized by the computer-executable code book generating method that is part of the process of obtaining the.

Images