JP2008145504A - Signal feature extracting method, signal feature extracting device, signal feature extracting program, and recording medium thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform signal feature extraction which is robuster than before by selecting a feature based upon a standard which is robuster than before and providing an area which is not a target of feature calculation. <P>SOLUTION: In a signal feature extraction device 1, a mask storage unit 20 stores a plurality of masks previously given identifiable mask numbers. An input unit 10 inputs a signal. A basic feature extracting unit 11 extracts basic feature information comprising a vector from the input signal. An in-mask maximum/minimum calculation unit 12 reads a mask out of the mask storage unit 20 and calculates a maximum value and a minimum value from a partial area, where the maximum value is calculated, and a partial area, where the minimum value is calculated, determined by the read-mask according to the extracted basic feature information. A difference calculation unit 13 calculates the difference between the calculated maximum value and minimum value. A mask selection vector quantizing unit 14 determines the mask number of a mask having the largest calculated differencde as a quantized value of an extraction target area of the basic feature information. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention provides, for example, a signal feature extraction method, a signal feature extraction apparatus, a signal feature extraction program, and a recording thereof for extracting a feature used for specifying a signal from a signal in audio signal detection and video signal detection. It relates to the medium.

Conventionally, as a signal feature extraction method, a method based on vector clustering, such as vector quantization based on an LBG algorithm, is known. Further, a method of extracting only a characteristic part of a signal, such as the method described in Patent Document 1, is also known.
International Publication No. 2006/006528 Pamphlet

  However, the method based on vector clustering has a problem that it is easily influenced by feature variations. Further, the method described in Patent Document 1 has a problem that there are few codes that can be expressed. In addition, the method based on vector clustering and the method described in Patent Document 1 have a problem that it is difficult to obtain a stable feature because the value of the feature vector varies depending on the sampling timing.

  The present invention has been made to solve the above-described problems, and its purpose is to select a feature based on a criterion that is more robust than in the past and to provide a signal that is more robust than in the past by providing an area that is not subject to feature calculation. It is an object to provide a signal feature extraction method, a signal feature extraction device, a signal feature extraction program, and a recording medium thereof that enable feature extraction.

  In order to solve the above problem, the present invention is a signal feature extraction method for extracting a feature used for extraction of a specific signal from an input signal, and a plurality of identification information that can be identified in advance is provided. A storage process for storing a mask in a mask storage means, an input process for inputting a signal, a basic feature extraction process for extracting basic feature information consisting of a vector from the input signal, and reading and extracting the mask from the mask storage means In-mask maximum / minimum calculation process for calculating the maximum value and the minimum value from the partial area for calculating the maximum value determined by the read mask and the partial area for calculating the minimum value, respectively, and the calculated minimum value And the difference calculation process for calculating the difference between the maximum value and the identification information of the maximum mask of the calculated difference as the quantization value of the extraction target region of the basic feature information, A signal feature extraction method which comprises a mask selection vector quantization process to output as a feature of the signal information including the quantized values for items.

  Further, the present invention is the above-described invention, wherein the mask stored in the mask storage means is a partial area for calculating a maximum value, a partial area for calculating a minimum value, and calculation of the maximum value and the minimum value. It is characterized in that partial areas not subject to the above are combined.

  Further, the present invention is the invention described in the above, wherein the partial area not to be calculated and the maximum value are set so that the partial area for calculating the maximum value and the partial area for calculating the minimum value are not adjacent to each other. The partial area to be calculated and the partial area to calculate the minimum value are combined.

  Further, in the present invention described above, when the basic feature information and the mask are indicated by a plurality of partial areas whose unit is a quadrangle in a two-dimensional plane, the present invention provides a partial area for calculating the maximum value; The partial area that is not subject to the calculation, the partial area that calculates the maximum value, and the minimum value are calculated so that the partial area that calculates the minimum value is not adjacent in at least four neighborhoods of the sides of the rectangle. And a partial region to be combined.

  Further, in the present invention described above, when the basic feature information and the mask are represented by a plurality of partial areas having a rectangular parallelepiped as a unit in a three-dimensional space, the partial area for calculating the maximum value is The partial area not to be calculated, the partial area for calculating the maximum value, and the minimum value are calculated so that the partial area for calculating the minimum value is not adjacent at least in the vicinity of 6 with respect to the surface of the rectangular parallelepiped. And a partial region to be combined.

  In addition, the present invention is a signal feature extraction device that extracts a feature used for extracting a specific signal from an input signal, and stores a plurality of masks to which identification information that can be identified in advance is provided. Means, input means for inputting a signal, basic feature extraction means for extracting basic feature information consisting of a vector from a signal input by the input means, a mask is read from the mask storage means, and the basic feature extraction means is In-mask maximum / minimum calculation means for calculating a maximum value and a minimum value respectively from a partial area for calculating a maximum value determined by the read mask and a partial area for calculating a minimum value of basic feature information to be extracted; Difference calculation means for calculating the difference between the minimum value and the maximum value calculated by the maximum / minimum calculation means, and identification information of the maximum mask of the difference calculated by the difference calculation means A signal feature comprising: a mask selection vector quantization unit configured to output information including a quantization value of the signal as a quantization value of the extraction target region of the basic feature information as a feature of the signal. It is an extraction device.

  Further, according to the present invention, a mask of a plurality of masks to which identification information that can be identified in advance is added to a computer of a signal feature extraction apparatus that extracts a feature used for extracting a specific signal from an input signal. A storage process to be stored; an input process to input a signal; a basic feature extraction process to extract basic feature information including a vector from the input signal; and a mask read out from the mask storage means, and the extracted basic feature information The maximum / minimum calculation process in the mask for calculating the maximum value and the minimum value respectively from the partial area for calculating the maximum value determined by the read mask and the partial area for calculating the minimum value, and the difference between the calculated minimum value and maximum value The difference calculation process to be calculated and the identification information of the maximum mask of the calculated difference as the quantization value of the extraction target area of the basic feature information, and the signal Information including the quantized value is a signal feature extraction program for executing a mask selection vector quantization process to output as a feature of the signal.

  Further, according to the present invention, a mask of a plurality of masks to which identification information that can be identified in advance is added to a computer of a signal feature extraction apparatus that extracts a feature used for extracting a specific signal from an input signal. A storage process for storing, an input process for inputting a signal, a basic feature extraction process for extracting basic feature information consisting of vectors from the input signal, a mask read out from the mask storage means, and the extracted basic feature information Maximum / minimum calculation process in the mask for calculating the maximum value and the minimum value from the partial area for calculating the maximum value determined by the read mask and the partial area for calculating the minimum value, and the difference between the calculated minimum value and maximum value The difference calculation process for calculating the difference and the identification information of the largest mask of the calculated difference as the quantized value of the extraction target area of the basic feature information, and the signal The information including the quantized value is a computer-readable recording medium recording a signal feature extraction program for executing the mask selection vector quantization process to output as a feature, the said signal.

  According to this invention, the signal feature extraction method stores a plurality of masks to which identification information that can be identified in advance is stored in the mask storage means, and extracts basic feature information including vectors from the input signal. The mask is read from the mask storage means, and the maximum value and the minimum value are calculated from the partial area for calculating the maximum value and the partial area for calculating the minimum value determined by the read mask of the extracted basic feature information. The difference between the minimum value and the maximum value is calculated, the identification information of the mask with the maximum calculated difference is set as the quantization value of the extraction target area of the basic feature information, and the information including the quantization value of the signal is set as the feature of the signal I decided to output it. This makes it possible to perform robust feature extraction at a point where a difference between features in a signal such as a corner is large by defining a partial region for calculating the maximum value and a partial region for calculating a minimum value in the mask. Is possible.

  According to the present invention, the mask stored in the mask storage means is a combination of a partial area for calculating the maximum value, a partial area for calculating the minimum value, and a partial area for which the maximum value and the minimum value are not calculated. It was decided that As a result, it is possible to perform feature extraction while reducing the influence of value fluctuation caused by a sampling timing shift at a location where a difference between features in a signal such as a corner is large.

  Further, according to the present invention, the partial area that is not subject to calculation, the partial area that calculates the maximum value, and the minimum value are calculated so that the partial area that calculates the maximum value and the partial area that calculates the minimum value are not adjacent to each other. To be combined with the partial area. As a result, it is possible to perform feature extraction while reducing the influence of value fluctuation caused by a sampling timing shift at a location where a difference between features in a signal such as a corner is large.

  Further, according to the present invention, when the basic feature information and the mask are indicated by a plurality of partial areas each having a quadrangle in the two-dimensional plane, the partial area for calculating the maximum value and the partial area for calculating the minimum value However, the partial area that is not subject to calculation, the partial area that calculates the maximum value, and the partial area that calculates the minimum value are combined so that they are not adjacent in the vicinity of at least four sides of the quadrangular side. As a result, in the basic feature information indicated by the power values in the time direction and the frequency direction such as acoustic signals, the fluctuation of the value due to the sampling timing shift at the location where the difference between the features in the signal such as the corner is large. It is possible to perform feature extraction while reducing the influence.

  Further, according to the present invention, when the basic feature information and the mask are indicated by a plurality of partial areas having a rectangular parallelepiped as a unit in the three-dimensional space, the partial area for calculating the maximum value and the partial area for calculating the minimum value However, the partial area that is not subject to calculation, the partial area that calculates the maximum value, and the partial area that calculates the minimum value are combined so that they are not adjacent in the vicinity of at least 6 in the plane of the rectangular parallelepiped. As a result, in the basic feature information indicated by the power values in the time direction and the frequency direction of the vertical and horizontal axes such as video signals, the value due to the sampling timing shift at the point where the difference between the features in the signal such as the corner is large. It is possible to perform feature extraction while reducing the influence of fluctuations.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
In the present embodiment, for example, in an enormous audio and video database, CM (Commercial Message), programs, and music signal fragments included in FM (Frequency Modulation) broadcast audio signals and television video signals are used as keys. The present invention relates to the implementation of a signal feature extraction method that detects the same signal at high speed, detects the location of the signal, and identifies the same content.

FIG. 1 is a schematic block diagram showing a signal feature extraction apparatus 1 according to the present embodiment.
The signal feature extraction apparatus 1 includes an input unit 10, a basic feature extraction unit 11, an in-mask maximum / minimum calculation unit 12, a difference calculation unit 13, a mask selection vector quantization unit 14, and a mask storage unit 20. ing.

  For example, the input unit 10 is connected to an output terminal of an audio device, a video device, or the like, and inputs a signal for feature extraction to the basic feature extraction unit 11. The basic feature extraction unit 11 extracts basic feature information indicated by a vector from the input signal. In the basic feature information, the mask storage unit 20 stores in advance a plurality of masks that specify areas in which the maximum / minimum calculation unit 12 in the mask calculates the maximum value and the minimum value.

  FIG. 2 is a diagram showing an example of a mask stored in the mask storage unit 20, where a mask number is assigned in advance to each mask, and in each mask, the area of the diagonal line rising to the right is the maximum in the mask. This indicates that the minimum calculation unit 12 is a target region for calculating the maximum value, and a diagonally downward slanting line region indicates that the maximum / minimum calculation unit 12 in the mask is a target region for calculating the minimum value. ing.

  The in-mask maximum / minimum calculation unit 12 reads out the mask from the mask storage unit 20, and specifies the partial region of the basic feature information specified for each read mask, that is, the target region for calculating the maximum value and the target for calculating the minimum value. The maximum value and the minimum value are calculated from the area. The difference calculation unit 13 calculates a difference between the minimum value and the maximum value calculated for each mask by the in-mask maximum / minimum calculation unit 12, for example, a value obtained by subtracting the maximum value from the minimum value.

  The mask selection vector quantization unit 14 determines the number of the mask with the largest difference among the differences between the minimum value and the maximum value for each mask calculated by the difference calculation unit 13, that is, the partial region identified by the mask, that is, attention It is set as a quantized value of a location, and the quantized value is output as feature information.

(Operation of signal feature extraction device)
Next, a signal feature extraction method performed by the signal feature extraction apparatus 1 will be described with reference to the flowchart of FIG. In the following description, a case where an acoustic signal is applied will be described as an example of a target signal. In the following description, a variable for storing the maximum difference value used in the comparison in step S6 in association with the mask number and a variable for storing the vector quantization value are determined in advance in the storage area. Assume that “NULL” is set as the initial value of each.

  First, the basic feature extraction unit 11 extracts basic feature information from the acoustic signal input from the input unit 10. Specifically, when the acoustic signal is sampled at the frequency F [Hz], the P point of the acoustic signal is subjected to Fourier transform, and P including the power value in each band of 0 to F ′ [Hz]. 'A dimensional vector, that is, a feature vector is extracted as basic feature information. Then, the basic feature extraction unit 11 extracts this feature vector every S seconds (step S1).

  For example, when an acoustic signal is composed of 8192 points (about 1 second) sampled at a frequency of 8000 Hz, this is Fourier-transformed to produce one frame with 4096 values consisting of power values in each band of 0 to 4000 Hz. This frame is extracted every 0.1 second.

  When the acoustic signal after Fourier transform is extracted every S seconds, the basic feature information is represented as a power value in a time-frequency space as shown in FIG. In FIG. 4, it is shown that the power value is different between the white area without dark shading and light shading, the thin shading area, and the dark shading area. At this time, as shown in FIG. 4, the basic feature extraction unit 11 divides the time-frequency space into cells with a square as a unit, and calculates an average value of power values for each divided region. FIG. 5 is a diagram illustrating a result of calculating an average value of power values for each cell with respect to the basic feature information of FIG.

  Next, when the basic feature information indicated by the power value as shown in FIG. 5 is input from the basic feature extraction unit 11 as shown in FIG. It is determined whether or not the region of the target location for which the value and the minimum value are to be calculated exists in the basic feature information (step S2). For example, in FIG. 4, the maximum value and the minimum value are calculated by shifting the 5 × 5 region of interest region by one square in the time direction and the frequency direction. At this time, in the input basic feature information, if the 5 × 5 region of interest to be calculated overlaps the region of the basic feature information, it is determined that the region of interest exists in the basic feature information. (Step S <b> 2: YES), when the maximum value and the minimum value are calculated in all the regions of interest existing in the basic feature information region, it is determined that the region of interest does not exist in the basic feature information. (Step S2: NO).

  Next, if the in-mask maximum / minimum calculation unit 12 determines that the region of interest exists, it determines whether or not the next mask exists in the mask storage unit 20 (step S3). For example, if there are eight masks shown in FIG. 2 in the mask storage unit 20, eight masks are read from the mask storage unit 20 one by one for one region of interest. At this time, if up to the seventh mask is read out for one region of interest, it is determined that the next mask exists (step S3: YES), and all the eight masks are read out. If it is determined, it is determined that the next mask does not exist (step S3: NO).

  For example, when the mask maximum / minimum calculation unit 12 reads the mask of mask number 1 in FIG. 2 as the first mask (step S3: YES), the mask read out in the region of interest as shown in FIG. Place. FIG. 6 is an enlarged view of the 5 × 5 region of interest in FIG. As described above, the mask maximum / minimum calculation unit 12 indicates that a region with a diagonal line rising to the right is a target region for calculating a maximum value, and a region with a diagonal line to the right is a maximum / minimum calculation within a mask. It shows that the part 12 is a target area for calculating the minimum value. In the example of FIG. 6, the in-mask maximum / minimum calculation unit 12 calculates 0.7 as the maximum value and calculates 3.9 as the minimum value.

  Next, the difference calculation unit 13 calculates a difference between the minimum value and the maximum value calculated by the in-mask maximum / minimum calculation unit 12, and the calculated difference and the in-mask maximum / minimum calculation unit 12 calculate the difference between the maximum value and the minimum value. The mask number used for the calculation is input to the mask selection vector quantization unit 14 (step S5). In the example of FIG. 6, 3.9−0.7 = 3.2 is calculated, “1” read first is selected as the mask number, and the calculated difference “3.2” and the mask number are selected. “1” is input to the mask selection vector quantization unit 14.

  Next, the mask selection vector quantization unit 14 compares the difference value input from the difference calculation unit 13 with the value stored in the variable of the maximum difference value, and calculates the difference input from the difference calculation unit 13. It is determined whether or not the value exceeds the value stored in the variable of the maximum difference value (step S6). Here, when “NULL” is stored in the variable of the maximum difference value, the value of the difference input from the difference calculation unit 13 exceeds the value stored in the variable of the maximum difference value. Judge as something. That is, in the determination process, in the first determination in the example of FIG. 6, the input difference value “3.2” is stored in the variable of the maximum difference value, and the variable of the mask number is input. The mask number “1” is stored.

  On the other hand, if it is determined that the difference value input from the difference calculation unit 13 does not exceed the value stored in the variable of the maximum difference value, the process returns to step S3, and the subsequent masks after step S4. The process will be repeated.

  When it is determined that the difference value input from the difference calculation unit 13 exceeds the value stored in the variable of the maximum difference value (step S6: YES), the mask selection vector quantization unit 14 The value stored in the mask number variable is stored in the vector quantization value variable, and the mask quantization value is updated (step S7). When the calculated difference “3.2” and the mask number “1” are stored in the variable of the maximum difference value and the variable of the mask number, respectively, “1” is stored as the vector quantization value.

  Here, the configuration of the mask shown in FIG. 2 used when detecting the corner (corner) in the direction of the arrow shown in the region selected as the target location from the basic feature information shown in FIG. 4 will be described. . The mask shown in FIG. 2 includes masks of mask numbers 2, 3, and 4 obtained by rotating the mask of mask number 1 by 90 degrees clockwise, a calculation target area of the maximum value of mask number 1, and a calculation target of the minimum value. The mask of mask number 5, which is an inverted region, and the masks of mask numbers 6, 7, 8 obtained by rotating the mask of mask number 5 by 90 degrees clockwise. As a feature of this mask, by providing an area that is not to be calculated between the maximum value calculation target area and the minimum value calculation target area, the maximum value calculation target area and the minimum value calculation target in the vicinity of 8 in a quadrangular side. This is to reduce the influence of fluctuations in the power value due to characteristic sampling timing shifts such as the corners shown in FIG.

  When detecting a corner, the maximum value calculation target area or the minimum value calculation target area is a concave shape or a convex shape, and the center of the attention point is either the maximum value calculation target area or the minimum value calculation target area. Good results can be obtained by setting the mask so as to belong to.

  FIG. 7 shows the respective results when the masks of mask numbers 1 to 8 shown in FIG. 2 are applied to the region of interest shown in FIG. 5 and the processing from step S3 to step S7 is performed. FIG. At this time, the maximum value of the difference value is “3.2” when the mask number is “1”, and the variable value of the vector quantization value after the last step S7 is “1”. As a result, the mask selection vector quantization unit 14 sets “1” to the coordinates of the center of the point of interest as shown in FIG. The vector quantization value calculated in the basic feature information is output as the feature of the signal.

  FIG. 9 is a diagram for illustrating the effect of the signal feature extraction apparatus 1 in the configuration of the above embodiment. In the example of FIG. 9, the maximum value calculation target region is obtained when the 5 × 5 mask shown in FIG. 2 is used at the 5 × 5 attention point and the 3 × 3 attention point that have the same center at the attention point. When using two 5 × 5 masks that do not have a non-calculation area between the minimum value calculation target area and the minimum value calculation target area, and not between the maximum value calculation target area and the minimum value calculation target area The calculation results of the respective maximum difference values when a 3 × 3 mask having no area is used are shown. The maximum difference value when the mask shown in FIG. 2 is used is calculated as 3.2 as described above. When two 5 × 5 masks having no non-calculation area between the maximum value calculation target area and the minimum value calculation target area are used, the maximum difference values of 0.3 and 2.0 are respectively obtained. When a 3 × 3 mask that is calculated and does not have a non-calculation area between the maximum value calculation target area and the minimum value calculation target area is used, a maximum difference value of −0.6 is calculated. .

  From this result, it can be seen that the mask size is larger when the 5 × 5 mask is used than when the 3 × 3 mask is used, and the intensity indicating the presence of corners can be increased by the value. Further, in a 5 × 5 mask and a 3 × 3 mask that do not have a region not to be calculated, the difference value between the minimum value and the maximum value becomes small, and the intensity indicating the presence of a corner is weakened by the value. As a result, the use of a mask having a region that is not a calculation target can express the presence of a corner with a stronger intensity than when a mask that does not have a calculation target is used, and more stably. It can be shown that a corner can be detected.

  In addition, from this, by using the signal feature extraction apparatus 1 of the present embodiment, by defining a region that is not a feature calculation target, a robust feature can be used by using a location where the difference between features in the signal is large. Can be detected, and features can be extracted based on criteria more robust than the above-described conventional technique.

  In the configuration of the above embodiment, it is not necessary to use all of the vector quantization values, and only the upper value of the larger absolute value of the difference value between the minimum value and the maximum value is selected in the peripheral region of the target location. In other words, it is possible to extract features in a sparse state.

  The mask shown in FIG. 2 in the embodiment described above has a shape for detecting corners in four directions. For example, when detecting a corner in eight directions, the shape shown in FIG. The eight masks may be added and applied.

  The mask shown in FIG. 2 is arranged so that the maximum value calculation target area and the minimum value calculation target area are not adjacent in the vicinity of 8 as described above, but in the vicinity of 4 as shown in FIG. You may make it arrange | position so that it may not adjoin.

  In the above embodiment, the target signal is described by taking the acoustic signal as an example. However, the present invention is not limited to the configuration of the above embodiment, and for example, the target signal may be a reduced image of the video signal. May be. In this case, for example, when the video signal is a television broadcast signal, when a video signal of about 15 seconds is searched from the television broadcast signal, first, an image of one frame of the television broadcast signal is X Equally divided and vertically divided into Y, for a total of X × Y areas. Then, the pixel values are averaged for each of RGB in each region. Then, an image composed of 3 × X × Y values composed of the average pixel values of RGB in each region is set as a reduced image per frame. For example, an image of one frame is divided into 16 equal parts horizontally and divided into 12 equal parts vertically to provide 192 areas, and pixel values are averaged for each of RGB in each area to generate a reduced image for one frame. In this case, the reduced image includes 576 values.

  The feature extraction method described above can be applied by generating the reduced image for each frame of the video signal and using the generated reduced image superimposed in the time direction for each RGB as basic feature information. At this time, the mask expanded in three dimensions is used. For example, FIG. 12 is a diagram showing one three-dimensional mask, and a three-dimensional mask is configured by overlapping masks 1-1 to 1-5 in the time axis direction.

  FIG. 13 shows a state in which two-dimensional reduced images superimposed in the time axis direction are arranged in a three-dimensional space, and a three-dimensional mask in which masks 1-1 to 1-5 are overlaid is placed at a point of interest. FIG. Similarly to the configuration of the above-described embodiment, the mask is shifted in the vertical axis, horizontal axis, and time axis directions in the basic feature information for each RGB to calculate the minimum value and the maximum value. A feature can be extracted by calculating a vector quantization value for each point of interest based on the difference.

  In addition, in the mask constituted by the mask 1-1 to the mask 1-5 shown in FIG. 12, the adjacent maximum value calculation target region and the minimum value calculation target region are arranged in the vicinity of 26 so as not to be adjacent to each other. The dimension mask can also be extended to a mask that detects 26 directions, or can be a mask that is not adjacent in the vicinity of 6. Further, it can be applied to a three-dimensional or more multi-dimensional mask.

  Also, a program for realizing each step shown in FIG. 3 is recorded on a computer-readable recording medium, and a program for realizing the function of the signal feature extraction device shown in FIG. A program for recording on a medium and for realizing the function of the processing unit in the signal feature extraction apparatus shown in FIG. 1 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is stored in a computer system. The signal feature extraction processing may be performed by reading and executing. Here, the “computer system” may include an OS and hardware such as peripheral devices.

  Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

  Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

It is the block diagram which showed the internal structure of the signal feature extraction apparatus by this embodiment. It is the figure which showed the mask memorize | stored in the mask memory | storage part in the same embodiment. It is the flowchart which showed the process of the signal feature extraction apparatus in the embodiment. It is the figure which showed an example of the signal input in the same embodiment. It is the figure which showed the example by which the mask was applied to the basic feature information output from the basic feature extraction part in the same embodiment, and the said basic feature information. It is the figure which showed the calculation object of the maximum value by the maximum / minimum calculation part in a mask in the same embodiment, and a minimum value. It is a figure for demonstrating the process of the maximum / minimum calculation part in a mask in the embodiment. It is the figure which showed the result as which the quantization by the mask selection vector quantization part in the same embodiment was performed. It is the figure which showed the comparison at the time of applying the processing result by the signal feature extraction method in the embodiment, and another mask. It is the figure which showed the example (the 1) of the other mask applicable to the signal feature extraction apparatus in the embodiment. It is the figure which showed the example (the 2) of the other mask applicable to the signal feature extraction apparatus in the embodiment. It is the figure which showed the example of the mask applied when a video signal is input into the signal feature extraction apparatus in the embodiment. It is the figure which showed the example of application of the mask at the time of inputting a video signal into the signal feature extraction device in the embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Signal feature extraction apparatus 10 Input part 11 Basic feature extraction part 12 Maximum / minimum calculation part in mask 13 Difference calculation part 14 Mask selection vector quantization part 20 Mask memory | storage part

Claims (8)

A signal feature extraction method for extracting features used for extracting a specific signal from an input signal,
A storage process of storing a plurality of masks to which identification information that can be identified in advance is stored in a mask storage unit;
An input process for inputting a signal;
A basic feature extraction process for extracting basic feature information consisting of vectors from the input signal;
In the mask for calculating the maximum value and the minimum value from the partial area for calculating the maximum value determined by the read mask and the partial area for calculating the minimum value of the basic feature information extracted from the mask storage means. Maximum and minimum calculation process;
A difference calculation process for calculating a difference between the calculated minimum value and maximum value;
A mask selection vector quantization process for outputting the identification information of the mask with the largest calculated difference as a quantization value of the extraction target region of the basic feature information, and outputting information including the quantization value of the signal as a feature of the signal; ,
A signal feature extraction method characterized by comprising: The mask stored in the mask storage means is
The signal feature extraction according to claim 1, wherein the partial area for calculating the maximum value, the partial area for calculating the minimum value, and the partial area for which the maximum value and the minimum value are not calculated are combined. Method. The partial area not to be calculated, the partial area for calculating the maximum value, and the partial area for calculating the minimum value so that the partial area for calculating the maximum value and the partial area for calculating the minimum value are not adjacent to each other. The signal feature extraction method according to claim 2, wherein When the basic feature information and the mask are represented by a plurality of partial areas each having a quadrangle on a two-dimensional plane, the partial area for calculating the maximum value and the partial area for calculating the minimum value are the quadrangle. The partial area not to be calculated, the partial area for calculating the maximum value, and the partial area for calculating the minimum value are combined so as not to be adjacent in at least four neighborhoods of the side of Item 3. The signal feature extraction method according to Item 2. When the basic feature information and the mask are indicated by a plurality of partial areas having a rectangular parallelepiped as a unit in a three-dimensional space, the partial area for calculating the maximum value and the partial area for calculating the minimum value are the rectangular parallelepiped. The partial area not to be calculated, the partial area for calculating the maximum value, and the partial area for calculating the minimum value are combined so as not to be adjacent at least in the vicinity of 6 of the surface. Item 3. The signal feature extraction method according to Item 2. A signal feature extraction device for extracting features used for extraction of a specific signal from an input signal,
Mask storage means for storing a plurality of masks to which identification information that can be identified in advance is provided;
An input means for inputting a signal;
Basic feature extraction means for extracting basic feature information consisting of vectors from a signal input by the input means;
A mask is read from the mask storage means, and a maximum value and a minimum value are respectively obtained from a partial area for calculating a maximum value determined by the read mask and a partial area for calculating a minimum value of basic feature information extracted by the basic feature extraction means. In-mask maximum / minimum calculation means for calculating a value;
A difference calculating means for calculating a difference between a minimum value and a maximum value calculated by the maximum / minimum calculating means in the mask;
The mask selection for outputting the information including the quantization value of the signal as the feature of the signal, using the identification information of the mask with the maximum difference calculated by the difference calculation means as the quantization value of the extraction target region of the basic feature information Vector quantization means;
A signal feature extraction apparatus comprising: In the computer of the signal feature extraction device for extracting features used for extraction of a specific signal from the input signal,
A storage process of storing a plurality of masks to which identification information that can be identified in advance is stored in a mask storage unit;
An input process for inputting a signal;
A basic feature extraction process for extracting basic feature information consisting of vectors from the input signal;
In the mask for calculating the maximum value and the minimum value from the partial area for calculating the maximum value determined by the read mask and the partial area for calculating the minimum value of the basic feature information extracted from the mask storage means. Maximum and minimum calculation process;
A difference calculation process for calculating a difference between the calculated minimum value and maximum value;
A mask selection vector quantization process for outputting the identification information of the mask with the largest calculated difference as a quantization value of the extraction target region of the basic feature information, and outputting information including the quantization value of the signal as a feature of the signal; ,
Signal feature extraction program for executing In the computer of the signal feature extraction device for extracting features used for extraction of a specific signal from the input signal,
A storage process of storing a plurality of masks to which identification information that can be identified in advance is stored in a mask storage unit;
An input process for inputting a signal;
A basic feature extraction process for extracting basic feature information consisting of vectors from an input signal;
In the mask for calculating the maximum value and the minimum value from the partial area for calculating the maximum value determined by the read mask and the partial area for calculating the minimum value of the basic feature information extracted from the mask storage means. Maximum and minimum calculation process;
A difference calculation process for calculating a difference between the calculated minimum value and maximum value;
A mask selection vector quantization process for outputting the identification information of the mask with the largest calculated difference as a quantization value of the extraction target region of the basic feature information, and outputting information including the quantization value of the signal as a feature of the signal; ,
The computer-readable recording medium which recorded the signal feature extraction program for performing this.

Images