JP2010224481A - Device for detection of similar section - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a similar section detection technology for searching a similar position at high speed, and for suppressing excessive detection or short of search. <P>SOLUTION: A feature value extraction unit 122 extracts a feature value string X from an image signal Vx for which a first input receiving unit 120 receives input, and extracts a feature value string Y from an image signal Vy for which a second input receiving unit 121 receives input. A similar position detection unit 123 time sequentially collates the feature value string X and the feature value string Y, and detects, as a similar position, a position where their similarity is larger than a search threshold θ stored in a threshold memory 124. A detection result output unit 125 binarizes a detection result r of the similar position detection unit 123 and outputs the binarized result r. A similar position distribution calculation unit 126 calculates a distribution of the number of the similar positions by using the detection result r. A threshold update section 127 updates the search threshold θ stored in a threshold memory 124, when a similar position distribution satisfies a predetermine distribution condition. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to optimization of a threshold value of a similar section detection device.

  For example, a technique for facilitating viewing by adding an index to a video by detecting a similar section from time series signals of two videos of the same program broadcast on different days has been proposed (for example, Patent Document 1). reference). A similar section is a set of consecutive similar positions. Searching for a similar section from two time-series signals such as video is called partial signal search. In the partial signal search, when the search window Wx is set on the time series signal X and the search window Wy is set on the time series signal Y, the signal in the search window Wx and the signal in the search window Wy are collated. The similarity between the search window Wx and the search window Wy is calculated, and when the similarity exceeds the search threshold θ, it is determined that the signal in the search window Wx is similar to the signal in the search window Wy. There is a method for searching for similar sections. However, if such a search is exhausted, the amount of processing becomes enormous. For this reason, in recent years, a method has been proposed in which collation at a position where the similarity is clearly below the search threshold value θ is omitted to speed up the search (see, for example, Non-Patent Document 1). This method is called the RIFAS method. In the RIFAS method, the search can be performed faster as the search threshold θ is larger, but there is a trade-off relationship that the search omission is increased.

Japanese Patent No. 40215545

Takuichi Nishimura, et al., “Detection of Matching Intervals in Time Series Data by Active Search Method—Reference Section Free Time Series Active Search Method”, “Science Theory D-II, Vol. J84-D-II, No. 8, pp. 1826-1837, Aug. 2001.

  If the search threshold θ is set to the maximum value at which no search omission occurs, the search can be performed at the highest speed. However, conventionally, it has been difficult to set the search threshold θ to an appropriate value, and the search threshold θ has been determined empirically. For this reason, there has been a risk of excessive detection and search omission.

  The present invention has been made in view of the above, and an object of the present invention is to provide a similar section detection apparatus that can search for similar positions at high speed and suppress excessive detection and search omission.

  In order to solve the above-described problems and achieve the object, the present invention is a similar section detection device, comprising a first input receiving means for receiving an input of a first time-series signal, and a second time-series signal. A second input receiving means for receiving an input; an extracting means for extracting a first feature quantity from the first time-series signal; and extracting a second feature quantity from the second time-series signal; A storage unit for storing a threshold value, the first feature value and the second feature value are collated in time series, and a similarity position whose similarity is equal to or higher than the first threshold value is obtained. Detecting means for detecting, calculating means for calculating the distribution of the number of similar positions, and when the distribution of the number of similar positions satisfies a predetermined distribution condition, the number of detected similar positions is reduced. Updating means for updating the first threshold, the detecting means, the calculating means, and Serial By repeatedly operating the updating means, characterized in that it comprises a control means for obtaining a similarity section said a set of similar contiguous positions.

  According to the present invention, it is possible to search for a similar position at high speed, and it is possible to suppress excessive detection and search omission.

FIG. 1 is a diagram illustrating a hardware configuration of a similar section detection device 100 according to an embodiment. FIG. 2 is a diagram illustrating a functional configuration of the similar section detection device 100. FIG. 3 is a flowchart illustrating a procedure of similar section detection processing performed by the similar section detection apparatus 100. FIG. 4 is a diagram for explaining the principle of optimizing the search threshold value θ.

  Exemplary embodiments of a similar section detection device according to the present invention will be explained below in detail with reference to the accompanying drawings.

[First embodiment]
(1) Configuration As shown in FIG. 1, the similar section detection device 100 includes a main storage unit such as a ROM (Read Only Memory) 104 and a RAM (Random Access Memory) 105 that stores various data and various programs, and various data and various types. It has an external storage unit 107 such as an HDD (Hard Disk Drive) or CD (Compact Disk) drive device for storing programs, and a bus 108 for connecting them, and has a hardware configuration using a normal computer. Yes. In addition, the similar section detection apparatus 100 includes a display unit 103 that displays information, an operation unit 102 such as a keyboard and a mouse that accepts user instruction input, and a communication unit 106 that controls communication with an external device. Each is connected by radio.

  Next, in such a hardware configuration, various functions realized when the CPU 101 of the similar section detection device 100 executes various programs stored in the storage unit or the external storage unit 107 will be described with reference to FIG. explain. The similar section detection apparatus 100 includes a first input reception unit 120, a second input reception unit 121, a feature amount extraction unit 122, a similar position detection unit 123, a threshold value storage unit 124, and a detection result output unit 125. And a similar position distribution calculating unit 126 and a threshold value updating unit 127. First input reception unit 120, second input reception unit 121, feature amount extraction unit 122, similar position detection unit 123, detection result output unit 125, similar position distribution calculation unit 126, and threshold value update unit 127 is generated in the storage unit such as the RAM 105 when the CPU 101 executes the program. The threshold storage unit 124 is a storage area generated in the external storage unit 107 such as an HDD.

  The first input receiving unit 120 receives an input of a video signal Vx that represents a video captured every unit time in time series. The second input receiving unit 121 receives an input of a video signal Vy different from the signal input to the first input receiving unit 120. The threshold storage unit 124 stores the search threshold θ (0 ≦ θ ≦ 1).

  The feature amount extraction unit 122 extracts the feature amount sequence X including the feature amount per unit time by extracting the feature amount per unit time from the video signal Vx received by the first input reception unit 120. At the same time, the feature amount sequence Y including the feature amount per unit time is extracted by extracting the feature amount per unit time from the video signal Vy received by the second input reception unit 121. A method for extracting a feature string from a video signal is, for example, as follows. The feature amount extraction unit 122 divides one frame image captured per unit time into nine regions, obtains an average luminance for each region, and quantizes it in four steps (2 bits). In addition to this, the feature quantity extraction unit 122 also obtains the average luminance of one entire frame image and quantizes it in 16 steps (4 bits). Then, the feature quantity extraction unit 122 concatenates these average luminances and extracts a 22-bit feature quantity. In this manner, the feature amount extraction unit 122 extracts feature amounts for each unit time, and extracts those obtained by arranging them in time series as a feature amount sequence.

  The similar position detection unit 123 searches the similar position by comparing the feature amount sequence X and the feature amount sequence Y in time series, and the similarity is larger than the search threshold θ stored in the threshold storage unit 124. The position is detected as a similar position. Specifically, the similar position detection unit 123 sets the search window Wx on the feature quantity sequence X and sets the search window Wy on the feature quantity sequence Y, and the feature quantity sequence and the search window in the search window Wx. The feature amount sequence in Wy is collated to search for similar positions in the video signals Vx and Vy. The search window is a predetermined time range, and is set on the feature amount sequence, so that a feature amount sequence including a corresponding feature amount within the predetermined time range is a target of collation. When matching such a feature quantity sequence in the search window Wx with a feature quantity sequence in the search window Wy, the similar position detection unit 123 calculates the similarity of the feature quantity sequence between the search windows Wx and Wy. When the similarity is larger than the search threshold value θ stored in the threshold value storage unit 124, the position x of the search window Wx and the p (x, y) representing the position y of the search window Wy are similar. Detect as position. Then, the similar position detection unit 123 outputs the detected similar position and the similarity s. The similar position detection unit 123, for example, fixes the search for the similar position by sequentially moving the position y of the search window Wy one sample at a time from 0, and freely moves the search window Wx at the fixed position y. And repeatedly for each position y.

  The method for calculating the similarity of the feature amount sequence between the search windows Wx and Wy is, for example, as follows. The similar position detection unit 123 obtains the histogram Hx of the feature amount sequence in the search window Wx and the histogram Hy of the feature amount sequence in the search window Wy, and calculates the histogram overlap rate indicating the ratio of the histograms Hx and Hy as similarities. Calculate as If the number of bins (columns) constituting the histograms Hx and Hy is M and the elements of the histogram are hxi and hyi (1 ≦ i ≦ M), the histogram overlap rate s (Hx, Hy) is expressed by the following equation 1 It can ask for.

  In the present embodiment, as shown in the RIFAS method, the similar position detection unit 123 performs high-speed similar position detection by omitting collation at a position where the similarity is clearly below the search threshold value θ. To explore. In the RIFAS method described above, when the similarity s is obtained at one point in the search space, there is a region where the similarity does not exceed the search threshold θ, that is, there is a skippable region where the search can be skipped. It is shown. In the RIFAS method, the entire search space is laid out in this skippable area and the search proceeds. The skippable area is a square area inclined by 45 degrees, and w can be expressed by the following expression 2 when the length of the diagonal line is 2w.

  This equation 2 shows that if the search threshold θ is large, the skippable area is also large. Therefore, if the search threshold θ is large, the search can be speeded up. On the other hand, if it is too large, a similar position below the search threshold value θ cannot be detected, and a search omission occurs. In the present embodiment, the function of the similar position detection unit 123 enables a similar position to be searched at high speed. Further, the search threshold value θ is optimized so that the number of detected similar positions is reduced by the functions of the detection result output unit 125, the similar position distribution calculation unit 126, and the threshold value update unit 127 described below. By doing so, excessive detection and search omission can be suppressed.

  The detection result output unit 125 binarizes and outputs the detection result r of the similar position detection unit 123. That is, when the similar position p (x, y) is output as a result of matching between the feature amount sequence in the search window Wx and the feature amount sequence in the search window Wy, the detection result output unit 125 displays “r = 1. The detection result r is output as “”, and when the similar position is not output, the detection result r is output as “r = 0”.

  The similar position distribution calculation unit 126 calculates the distribution of the number of similar positions using the detection result r output from the similar position detection unit 123. Specifically, the similar position distribution calculation unit 126, for example, for each detection result r output in the search window Wx set for the search window Wy for each position y of the search window Wy, The number of 1 is calculated, and the average value N of the number of similar positions around the unit length is calculated.

  The threshold update unit 127 updates the search threshold θ when the similar position distribution calculated by the similar position distribution calculation unit 126 satisfies a predetermined distribution condition, and the updated value is stored in the threshold storage unit. 124 is stored. The predetermined distribution condition is, for example, that the average value N calculated by the similar position distribution calculation unit 126 exceeds the update threshold value α1. The update threshold value α1 is stored in advance in the external storage unit 107 such as an HDD. The search threshold value θ is updated, for example, by adding a constant d to the search threshold value θ every time it is updated. As a result, each time the update is performed, the search threshold θ increases by a constant d. The constant d is stored in advance in the external storage unit 107 such as an HDD. When the similar position detection unit 123 searches for a similar position using the search threshold θ updated in this way, the number of detected similar positions is smaller than when the search threshold θ before the update is used. Will reduce and overdetection will decrease.

  In the present embodiment, when the average value N exceeds the update threshold value α1, it is determined that excessive detection has occurred for the following reason. In a two-dimensional XY space in which the feature quantity sequence X is the X axis and the feature quantity sequence Y is the Y axis, the similar section is detected as a line segment with a slope of 1. On the other hand, excessive detection is detected as a line segment parallel to the X axis or the Y axis. Therefore, the similar position detector 123 detects the line segment parallel to the X axis by calculating the average value N of the number of similar positions per unit length for each position y, and this average value N and the update threshold value α1. This is because it can be determined whether or not excessive detection has occurred. In this embodiment, when it is determined that excessive detection has occurred, the threshold update unit 127 reduces the excessive detection, that is, the number of detected similar positions decreases. The search threshold θ is updated so that In this way, whenever the average value N calculated by the similar position distribution calculation unit 126 for each position y becomes larger than the update threshold value α1, the threshold value update unit 127 updates the search threshold value θ so that the search is performed. The threshold value θ is optimized, and the similar position detection unit 123 searches for a similar position using the optimized search threshold value θ. The similar position detection unit 123 repeats such processing until the excessive detection does not occur in the process of searching for similar positions, thereby optimizing the search threshold θ, and selecting a similar section that is a set of consecutive similar positions. get.

(2) Operation Next, a procedure of similar section detection processing performed by the similar section detection apparatus 100 according to the present embodiment will be described with reference to FIG. In step S <b> 1, the similar section detection device 100 receives an input of the video signal Vx by the function of the first input receiving unit 120 and receives an input of the video signal Vy by the function of the second input receiving unit 121. Then, the similar section detection device 100 extracts the feature amount sequence X from the video signal Vx and the feature amount sequence Y from the video signal Vy by the function of the feature amount extraction unit 122. In step S <b> 2, the similar section detection device 100 sets a search window Wx on the feature quantity sequence X and sets a search window Wy on the feature quantity sequence Y by the function of the similar position detection unit 123. The position y of the window Wy is initialized to 0. In step S <b> 3, the similar section detection device 100 uses the function of the similar position detection unit 123 to perform all the positions y, x in the two-dimensional XY space with the feature quantity sequence X as the X axis and the feature quantity sequence Y as the Y axis. To determine whether the search for similar positions is completed. If the determination result is affirmative, the process ends. If the determination result is negative, the position y of the search window Wy is set in the positive direction. Move one sample and go to step S4.

  In step S4, the similar section detection apparatus 100 fixes the position y of the search window Wy, and freely moves the position x of the search window Wx (for example, moves it in the positive direction by one sample. Alternatively, the search is performed by the RIFAS method. It moves to the required position.), The feature amount sequence in the search window Wx and the feature amount sequence in the search window Wy are collated to search for a similar position. At the time of this collation, the similar section detection device 100 calculates the similarity of the feature amount sequence between the search windows Wx and Wy, and the similarity is stored in the threshold storage unit 124. It is determined whether it is greater than θ. When the similarity is larger than the search threshold θ stored in the threshold storage unit 124, the similar section detection device 100 detects the position p (x, y) of the search windows Wx, Wy as a similar position. Then, the similar position and the similarity s are output. Further, the similar section detection device 100 outputs the detection result r of the similar position by the function of the detection result output unit 125. Note that using the RIFAS method described above, in step S4, the similar section detection device 100 omits collation at a position (skipable area) where the similarity is clearly below the search threshold θ.

  When the search for similar positions for all the positions x with respect to the position y is completed, in step S5, the similar section detection device 100 detects each of the positions x at the position y by the function of the similar position distribution calculation unit 126. The number of r having a value of 1 is calculated, and an average value N of the number of similar positions per unit length is calculated. In step S6, the similar section detection device 100 determines whether or not the average value N calculated in step S5 exceeds the update threshold value α1 by the function of the threshold value update unit 127. If the average value N exceeds the update threshold value α1, it means that over detection has occurred. In this case, in step S7, the similar section detection device 100 updates the search threshold θ by adding a constant d to the search threshold θ stored in the threshold storage unit 124, and updates the updated value. Is stored in the threshold storage unit 124. Thereafter, the process returns to step S3.

  Next, the principle of optimizing the search threshold θ in the present embodiment will be described with reference to FIG. Reference numeral 301 denotes search windows Wx and Wy set in a two-dimensional XY space having the feature quantity string X as the X axis and the feature quantity string Y as the Y axis, and similar positions detected by the similar position detection unit 123 in the XY space. Is represented as a set of sunspots. In the figure, a set of black dots is represented as a line segment having a slope of 1, a line segment parallel to the X axis, or a line segment parallel to the Y axis. Reference numeral 302 represents a transition of the number of similar positions per unit length calculated by the similar position distribution calculation unit 126 for each position on the Y axis. Reference numeral 303 represents a transition of the search threshold θ updated by the threshold update unit 127.

  First, the difference between the similar position and the true similar position due to excessive detection will be described. As described above, the similar section detection device 100 moves and fixes the position y of the search window Wy one sample at a time from 0, and searches the similar position by freely moving the search window Wx at the fixed position y. This process is repeated for each position y. As indicated by reference numeral 301, true similar positions are distributed around a line segment having an inclination of 1. On the other hand, similar positions due to excessive detection are distributed around a line segment parallel to the X axis or the Y axis. For example, if the video represented by the video signal Vy has a dark part, the dark part may be detected as a similar position as being similar to the dark part in the video represented by the video signal Vx. In addition, when frame images having dark portions are continuous in the video signal Vx, similar positions tend to be distributed around the line segment parallel to the X axis. Since the dark part itself does not exactly match, the similarity in the dark part due to overdetection is lower than the similarity in the true similar position. For this reason, if the search threshold value is appropriately set, the true similar position and the similar position due to excessive detection can be distinguished.

  Next, excessive detection of similar positions and search threshold optimization will be described. First, the similar section detection device 100 starts the search by setting the search threshold θ to a value θo lower than the optimum value θa assumed as the optimum value of the search threshold. That is, θo is stored in advance in the threshold storage unit 124 as an initial value of the search threshold θ. As described above, similar positions due to excessive detection are distributed around the line segment parallel to the X axis. Therefore, when the position of the search window Wy advances from 0 to y1, for example, as indicated by reference numeral 302, the position at the position y1 The number of similar positions per unit length about the X axis increases rapidly. When the average value of this number is N (y1), the similar section detection device 100 determines that excessive detection has occurred when the following expression 3 holds. α1 is the update threshold value described above.

  If the similar section detection device 100 determines that excessive detection has occurred, the similar section detection device 100 updates the search threshold by increasing the search threshold θ by a constant d. Assuming that the updated search threshold is θ ′, the following Expression 4 is established.

  The similar section detection apparatus 100 searches for a similar position by using the updated search threshold value θ ′ as the search threshold value θ. As the search threshold θ increases, the number of detected similar positions decreases. However, when the updated search threshold value θ is smaller than the optimum value, as the search proceeds, the number of similar positions per unit length of the X axis becomes the update threshold value α1 as the reference numeral 302 indicates. It can happen repeatedly. Therefore, when such a situation occurs, the similar section detection apparatus 100 performs a search threshold so that the number of detected similar positions is reduced by updating the search threshold θ again using Equation 4. Optimize the value θ. In this way, according to the present embodiment, the search threshold can be automatically set to the optimum value, and as a result, excessive detection and search omission can be suppressed, and a similar position can be searched at high speed. can do.

[Second Embodiment]
Next, a second embodiment of the similar section detection device will be described. In addition, about the part which is common in the above-mentioned 1st Embodiment, it demonstrates using the same code | symbol or abbreviate | omits description.

(1) Configuration In the present embodiment, a configuration is described in which an audio signal is handled as a time series signal and a similar section is detected from two audio signals. The functional configuration itself of the similar section detection apparatus 100 according to the present embodiment is substantially the same as that illustrated in FIG. The configuration according to the present embodiment is different from the first embodiment described above as follows. The first input receiving unit 120 receives an audio signal input every unit time, and the second input receiving unit 121 receives an audio signal input every unit time. The feature amount extraction unit 122 extracts a feature amount sequence including a feature amount per unit time by extracting a feature amount per unit time from the audio signal received by the first input reception unit 120, and The feature amount sequence including the feature amount for each unit time is extracted by extracting the feature amount for each unit time from the audio signal received by the second input receiving unit 121. A method for extracting a feature string from an audio signal is, for example, as follows. Note that the audio signal is, for example, a 16-bit PCM signal sampled at 11025 [Hz]. The feature amount extraction unit 122 has a secondary IIR digital filter, and a band pass centered on seven frequencies obtained by dividing the logarithm from 200 [Hz] to 3700 [Hz] into 6 parts using the secondary IIR digital filter. A filter group is configured. The secondary IIR digital filter may be realized by the CPU 101 executing a program stored in the ROM 104 or the external storage unit 107, or may be configured by a DSP or a sequential circuit. The band pass filter group is input with the audio signal received by the first input receiving unit 120 and the audio signal received by the second input receiving unit 121, and the feature amount extracting unit 122 is arranged at regular intervals with respect to time. Calculate the power. Since the number of channels is 7, seven powers P (i) (1 ≦ i ≦ 7) can be obtained by one sampling. The sampling interval is 29.97 [Hz]. Then, the feature amount extraction unit 122 vector-quantizes the power value P (i) to 2 bits using three threshold values to obtain B (i). The feature amount extraction unit 122 generates a 14-bit code by concatenating the vectors in the order of B (1) B (2) B (3) B (4) B (5) B (6) B (7), This is defined as a feature amount of the audio signal. As a threshold used in vector quantization, a value similar to the sound represented by the input sound signal is analyzed to obtain a value with the highest coding efficiency, and this value may be used as a fixed value.

(2) Operation Next, a procedure of similar section detection processing performed by the similar section detection apparatus 100 according to the present embodiment will be described. The procedure of the similar section detection process according to this embodiment is substantially the same as that shown in FIG. In step S <b> 1, the similar section detection device 100 receives an input of an audio signal by the function of the first input receiving unit 120 and receives an input of an audio signal by the function of the second input receiving unit 121. Then, the similar section detection device 100 extracts a feature amount sequence from each of the two audio signals that have been input by the function of the feature amount extraction unit 122. Step S2 and subsequent steps are the same as those in the first embodiment described above.

  According to the configuration as described above, even when searching for a similar position in the audio signal, the search threshold can be automatically set to the optimum value, and as a result, excessive detection and search omission are suppressed. It is possible to search for similar positions at high speed.

[Modification]
Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Moreover, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined. Further, various modifications as exemplified below are possible.

<Modification 1>
In each embodiment described above, various programs executed by the similar section detection device 100 may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. good. The various programs are recorded in a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, and a DVD (Digital Versatile Disk) in a file in an installable or executable format. The computer program product may be provided.

  Further, the first input receiving unit 120, the second input receiving unit 121, the feature amount extracting unit 122, the similar position detecting unit 123, the detection result output unit 125, the similar position distribution calculating unit 126, a threshold value The update unit 127 is generated in the storage unit such as the RAM 105 when the CPU 101 executes the program, but at least one of these may be configured by hardware such as a processor.

<Modification 2>
In each of the embodiments described above, the similar position detection unit 123 searches the similar position by fixing the search window Wy to the position y and freely moving the position x of the search window Wx. May be fixed at position x and the position y of the search window Wy may be freely moved to search for similar positions.

  In each of the above-described embodiments, the predetermined distribution condition for determining whether or not excessive detection has occurred is that the average value of the number of similar positions when the search window Wy is fixed is the update threshold value. As long as α1 is exceeded, any distribution condition may be used as long as the condition utilizes the property that similar positions due to overdetection are distributed around a line segment parallel to the X axis or the Y axis.

  For example, over-detection may be determined using a predetermined distribution condition that the rate of change in the number of similar positions exceeds the update threshold value α2. For example, the number of similar positions when the search window Wy is at the position y (i) is N (y (i)), and the number of similar positions when the search window Wy is at the position y (i + 1) is N (y (i + 1)). Then, the change rate A (i) of the similar position can be expressed by the following formula 5.

  It is assumed that over-detection and over-detection have occurred, assuming that a predetermined condition is satisfied when the rate of change A (i) of the similar position exceeds a predetermined update threshold value α2 and satisfies the relationship expressed by the following expression 6. Judgment is sufficient.

  Also, instead of counting the number of similar positions only on the line segment with the search window Wy fixed, a two-dimensional XY space with the feature quantity sequence X as the X axis and the feature quantity sequence Y as the Y axis is divided into 16 equal parts. Then, overdetection may be determined on the condition that the number N of similar positions in each equally divided region exceeds the update threshold value α3.

  In the case where the search is advanced while moving the search window Wy little by little in the two-dimensional XY space having the feature quantity row X as the X axis and the feature quantity row Y as the Y axis as in the above-described embodiments. Further, it is also possible to detect a line segment parallel to the X-axis or the Y-axis using the Hough transform, and to determine over-detection with a predetermined distribution condition that the length of the line segment exceeds the update threshold value α4. .

  Also, as in the above-described embodiments, the number of similar positions is stored while moving the search window Wy little by little, and a predetermined distribution condition is set based on the transition of the number of similar positions, and excessive Detection may be determined. For example, when the variance is obtained from the transition of the number of similar positions and this variance is below the update threshold value α5, it indicates that the similar positions are extremely densely distributed. If the condition is satisfied, it is determined that excessive detection has occurred.

<Modification 3>
In the first embodiment described above, in the RIFAS method, the region evaluated as the skippable region is managed by the column U of the valley position. Depending on whether or not the number of the valleys U exceeds the update threshold value α6, Excess detection may be determined. The principle of searching for a similar position by the RIFAS method will be briefly described below to explain what kind of valley is used in this method. In the RIFAS method, when similarity is obtained at a certain position in a two-dimensional XY space (referred to as a search space) in which the feature quantity sequence X is the X axis and the feature quantity sequence Y is the Y axis, Indicates that there is a region that cannot exceed the search threshold θ, and the search is speeded up by omitting verification in this region. This area is a skippable area. The skippable area is a square inclined by 45 degrees with the position where the similarity is obtained as the center, and the diagonal length of the square can be calculated from the similarity. In the RIFAS method, the partial signal search problem is formulated as a problem in which a two-dimensional search space is spread by skippable areas or similar positions. First, when the search window Wx is scanned and the search window Wx is scanned to evaluate whether all the positions are skippable areas or similar positions, the process of moving the search window Wy by one sample is repeated. At this time, it is necessary to store the shape of the boundary line between the skippable region or the region evaluated as a similar position and the unevaluated region. In the RIFAS method, when the skippable areas overlap, the intersection is stored, and when it is a similar position, the position is stored. These positions are smaller on the Y axis than the positions in other skippable areas and look like valleys. Therefore, these positions are called valleys. When the number of similar positions is small, the skippable area is large and the number of valleys is also small. Conversely, when there are many similar positions, the skippable area is small, and the number of valleys is large. That is, the number of valleys increases as the number of similar positions increases, and decreases as the number of similar positions decreases. Using this property, instead of directly counting the number of similar positions, the number of valleys is counted, and overdetection is determined on the condition that the number of valleys exceeds the update threshold value α6. Also good.

  According to the above configuration, since the number of valleys is known during the search process, a mechanism for counting the number of similar positions may not be newly added to the search process.

  In addition, as a predetermined distribution condition, a parameter value that is used when the similar position detection unit 123 detects a similar position and that increases as the number of similar positions increases exceeds the update threshold value α7. Excess detection may be determined.

  The target value for determining whether or not the update threshold is exceeded is the number of similar positions, the rate of change of the number of similar positions, the length of a line segment obtained by Hough transform, and parameters. It may be at least one of at least one maximum value, minimum value, average value, and variance.

<Modification 4>
In the first embodiment described above, the search threshold value θ is updated by adding a constant d to the search threshold value θ, but the search is performed so that the number of detected similar positions decreases. The present invention is not limited to this as long as the threshold value θ is updated. For example, a value obtained by dividing one or more constants e by the search threshold θ may be used as a new search threshold. The updated search threshold value θ ′ can be expressed by the following Equation 7.

  When the search threshold is far away from its optimal value, it is preferable that the update amount for updating the search threshold is large, and when the search threshold is not far from the optimal value. Therefore, it is preferable that the update amount is small. In the configuration of this modification, “θ ≦ 1”, so the update amount gradually decreases. Therefore, the search threshold can be brought closer to the optimum value as compared with updating the search threshold by adding a constant.

100 Similar Section Detection Device 101 CPU
102 Operation unit 103 Display unit 104 ROM
105 RAM
106 communication unit 107 external storage unit 108 bus 108 threshold update unit 120 first video input reception unit 121 second video input reception unit 122 feature amount extraction unit 123 similar position detection unit 124 threshold storage unit 125 detection result output unit 126 Similar position distribution calculation unit 127 Threshold update unit

Claims (6)

First input receiving means for receiving an input of a first time-series signal;
Second input receiving means for receiving an input of the second time-series signal;
Extraction means for extracting a first feature quantity from the first time series signal and extracting a second feature quantity from the second time series signal;
Storage means for storing a first threshold;
Detecting means for collating the first feature quantity and the second feature quantity in time series and detecting a similar position whose similarity is equal to or higher than the first threshold value;
Calculating means for calculating a distribution of the number of similar positions;
Updating means for updating the first threshold value so that the number of detected similar positions decreases when the distribution of the number of similar positions satisfies a predetermined distribution condition;
A similar section detection apparatus comprising: a control section that acquires a similar section that is a set of similar positions by repeatedly operating the detection section, the calculation section, and the update section. The predetermined distribution condition includes a two-dimensional space having the number of similar positions, the rate of change of the number of similar positions, the first feature quantity as a first axis, and the second feature quantity as a second axis. It is set based on at least one of the lengths of the line segments parallel to the first axis or the second axis obtained by straight line detection above.
The updating means includes
Determining means for determining whether or not the predetermined distribution condition is satisfied;
The similar section detection device according to claim 1, further comprising a threshold update unit configured to update the first threshold when the predetermined distribution condition is satisfied. The predetermined distribution condition is that the number of similar positions exceeds a second threshold, the rate of change of the number of similar positions exceeds a third threshold, and the length of the line segment is the first. At least one of exceeding the threshold of 4;
The similar section detection device according to claim 2, wherein the determination unit determines whether or not the predetermined distribution condition is satisfied. The predetermined distribution condition can be used to detect a similar position, and a parameter that increases with an increase in the number of the similar positions can be omitted, and verification of the first feature quantity and the second feature quantity can be omitted. It is set based on at least one of the number of valleys managing the area,
The updating means includes
Determining means for determining whether or not the predetermined distribution condition is satisfied;
The similar section detection device according to claim 1, further comprising a threshold update unit configured to update the first threshold when the predetermined distribution condition is satisfied. The predetermined distribution condition is at least one of the parameter exceeding a fifth threshold and the number of valleys exceeding a sixth threshold;
The similar section detection apparatus according to claim 4, wherein the determination unit determines whether or not the predetermined distribution condition is satisfied. The updating unit adds a constant to the first threshold value or newly sets a value inversely proportional to the first threshold value when the distribution of the number of similar positions satisfies a predetermined distribution condition. The similar section detection device according to claim 1, wherein the first threshold value is updated by setting the first threshold value.

Images