JP5028651B2 - Information processing apparatus and content analysis program - Google Patents

Description

  The present invention relates to an information processing apparatus and a content analysis program.

  In recent years, the storage capacity of storage devices has increased, and various contents can be acquired by broadcasting or distribution. Accordingly, there is a need for a device that can detect and index a predetermined section of content and reproduce only a section desired by the user. For example, in a recorder intended to record and replay a broadcast program, when receiving and recording a signal transmitted from a broadcast station, CM sections and highlight sections are detected and indexed. There is known a recording apparatus that can reproduce only a scene indexed as a scene that a user wants to see using information.

In this way, an apparatus that indexes a predetermined section from video / audio data calculates a predetermined feature amount by analyzing an audio signal or a video signal, and detects the predetermined section using the feature amount. For example, in Patent Document 1, music section detection is performed in which an orthogonal transform coefficient obtained by orthogonally transforming audio data digitized by the PCM method into a frequency domain is calculated, and a predetermined section is detected using a feature amount based on the orthogonal transform coefficient. A device has been proposed.
JP 2007-180669 A

  As in the invention described in Patent Document 1, in order to analyze a digitally compressed acoustic signal, it is necessary to calculate a feature amount after performing decoding processing on all frames.

  However, when such an analysis is performed with a low-spec device such as a portable terminal, the load of decoding processing for all frames is high. For this reason, for example, it is difficult to perform content analysis processing in which a mobile terminal receives and stores encoded content from one seg broadcasting or the like and performs analysis in real time. Even when content analysis processing is not performed in real time, battery consumption increases because decoding processing is performed for all frames.

  Moreover, when analyzing the content acquired by the mobile terminal through communication, a transmission error is likely to occur, and it is difficult to accurately detect the section.

  SUMMARY OF THE INVENTION An object of the present invention is to provide an information processing apparatus that can analyze content with low memory-saving and low computation while maintaining content analysis accuracy.

  In order to achieve the above object, an information processing apparatus according to the present invention is an information processing apparatus that classifies a predetermined section from encoded content, and decodes frames included in the content by thinning out at a predetermined thinning rate. A decoding means; a frame feature quantity calculating means for calculating a feature quantity of a frame decoded by the decoding means; and a frame feature quantity calculating means for discretizing a feature quantity of a segment consisting of a predetermined frame group. If the difference between the calculated feature value of the segment unit and the reference value is greater than or equal to a predetermined value, the feature value of the segment unit is used. The segment feature value is discretized and the difference between the calculated segment feature value and the reference value is less than the predetermined value. For example, the decoding means performs decoding by changing the frame thinning rate until the difference between the segment unit feature amount and the reference value is equal to or greater than a predetermined value or until the segment unit feature amount is calculated based on the feature amount of all frames. The segment feature quantity discretizing means for discretizing the segment feature quantity using the finally obtained segment unit feature quantity, and the discretization obtained by the segment unit feature quantity discretizing means It has the section classification means which classifies the section of the content using the feature amount of the segment.

  The content analysis program according to the present invention is a content analysis program for classifying a predetermined section from encoded content, the decoding function for thinning and decoding the frames included in the content at a predetermined thinning rate, and the decoding function A frame feature amount calculation function for calculating the feature amount of the frame decoded by the step S, and a feature of the frame calculated by the frame feature amount calculation function when the feature amount of a segment consisting of a predetermined frame group is discretized and calculated. If the difference between the calculated feature value of the segment unit and the reference value is equal to or greater than a predetermined value, the segment feature value is discrete using the segment unit feature value. The difference between the calculated feature value of the segment unit and the reference value is less than the predetermined value. If the difference between the segment unit feature amount and the reference value is equal to or greater than a predetermined value or the segment unit feature amount is calculated from the feature amount of all frames, the frame decimation rate is changed and decoding is performed by the decoding function. The segment feature discretization function that discretizes the segment feature quantity using the finally obtained segment unit feature quantity, and the discretized discretization obtained by the segment unit feature quantity discretization function It has a section classification function for performing section classification of content using the feature amount of the segment.

  ADVANTAGE OF THE INVENTION According to this invention, the information processing apparatus which can analyze a content by memory-saving low calculation can be provided, maintaining the analysis accuracy of a content.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings, taking as an example a mobile phone capable of receiving and recording one-segment broadcasting. However, the present invention is not limited to a mobile phone, and any device capable of reproducing multimedia contents may be used. For example, it may be a device that receives and broadcasts digital broadcasts such as a DVD recorder, or may be a playback device that plays back content acquired via the Internet.

  FIG. 1 is a block diagram showing a configuration of a mobile phone according to the present embodiment. Control of the entire mobile phone 1 is performed by the control unit 51, and the control unit 51 includes a CPU, a ROM, and a RAM. The mobile phone 1 also includes an antenna 52 for wireless communication and an antenna 58 for receiving one-segment broadcasting.

  The antenna 52 receives the radio wave output from the base station and converts it into an electrical signal. The RF unit 53 converts the frequency of the electrical signal input from the antenna 52 and outputs a baseband signal to the communication control unit 54. The communication control unit 54 modulates the baseband signal and separates the error-corrected data for each medium. For example, when a call voice is received, the audio data included in the received data is decoded and output to the voice input / output unit 55. For example, if the received data includes moving image data such as a videophone, the moving image data is decoded and output to the display unit 56.

  The antenna 58 receives the radio wave of the one-segment broadcasting, acquires the selected physical channel signal, converts it to an electrical signal, and outputs it to the RF unit 59. The RF unit 59 converts the received signal into a baseband signal and outputs it to the broadcast receiving unit 60. The broadcast receiving unit 60 performs AD conversion on the input signal, performs OFDM demodulation, performs error correction, and obtains a TS (Transport Stream) packet.

  When reproducing this received data, the ES (Elementary Stream) of Audio ES, Video ES, and Text ES is separated from the TS packet, and each ES is decoded. The decoded Audio ES is output to the audio input / output unit 55. VideoES and TextES are output to the display unit 56.

  On the other hand, when TS packets are stored, a predetermined section can be stored in the storage unit 57 while indexing from the broadcast content. The content processing unit 100 performs the content indexing process. For example, when the received one-segment broadcasting is recorded in the content processing unit 100, by viewing and storing the CM section while indexing, the CM section is viewed or only the CM section is viewed during viewing. It becomes possible.

  FIG. 2 is a block diagram illustrating a configuration of the content processing unit 100. The content processing unit 100 includes a demux unit 101, a transmission error detection filter unit 102, a decoding unit 103, a content analysis unit 104, a section classification unit 105, a TS data temporary storage unit 106, and an analysis result integration unit 107.

  The TS packet acquired by the broadcast receiving unit 60 is input to the Demux unit 101 and the TS data temporary storage unit 106. The TS data temporary storage unit 106 temporarily stores TS packets until the analysis result integration unit 107 finishes processing of the corresponding TS packet. In the present embodiment, the TS data temporary storage unit 106 is provided as a temporary storage destination of TS data. However, the TS data temporary storage unit 106 is not provided, and TS data is temporarily stored in a multimedia content DB unit 57a described later. It is also possible to make a configuration in which the data is stored.

  The demux unit 101 separates and extracts at least one ES (Elementary Stream) of AudioES, VideoES, and TextES from the input TS packet.

  The separated and extracted ES is subjected to transmission error detection by the transmission error detection filter unit 102 and decoding processing by the decoding unit 103. Then, the content analysis unit 104 analyzes the content using the decoded data input from the decoding unit 103.

  The section classification unit 105 performs processing for classifying a predetermined section from the content using the result analyzed by the content analysis unit 104.

  In the analysis result integration unit 107, the result analyzed by the section classification unit 105 and the data stored in the TS data temporary storage unit 106 are associated and integrated, output as multimedia content with metadata, and stored in the storage unit 57. This multimedia content with metadata is stored in the multimedia content DB unit 57a.

  With such a configuration, it is possible to analyze the content and attach the analysis result to the content as metadata. Hereinafter, the content analysis unit 104, the transmission error detection filter unit 102, and the section classification unit 105 will be described in detail.

(Content analysis unit 104)
FIG. 3 is a block diagram of the content analysis unit 104 that extracts feature amounts for performing section classification processing on content. The content analysis unit 104 includes an analysis control unit 104a, a segment unit feature amount extraction unit 104b, and a time duration feature amount extraction unit 104c. Either one or both of the segment unit feature amount extraction unit 104b and the time continuation feature amount extraction unit 104c operate depending on what kind of section classification processing is performed on the content. Further, the time continuation feature amount extraction unit 104c may be operated using the first feature amount extracted by the segment unit feature amount extraction unit 104b.

  The analysis control unit 104a outputs a decoding request to the decoding unit 103 so as to decode the frame with a predetermined thinning rate. When the decoded data is acquired from the decoding unit 103, the segment unit feature amount extraction unit 104b or the time continuation feature amount extraction is performed so that the section classification unit 105 operates the process for extracting the feature amount used for the section classification process. The unit 104c is controlled. When the analysis control unit 104a controls the segment unit feature amount extraction unit 104b to operate, the decode data acquired from the decode unit 103 is output to the segment unit feature amount extraction unit 104b and the segment unit feature amount extraction unit 104b. The first analysis requirement used for the feature quantity extraction process at is output. When the analysis control unit 104a controls the time continuation feature amount extraction unit 104c to operate, the decode data acquired from the decoding unit 103 is output to the time continuation feature amount extraction unit 104c and the time continuation feature amount extraction is performed. The second analysis requirement used for the feature amount extraction processing in the unit 104c is output.

  If the value calculated as the feature amount in the content analysis unit 104 is not waveform information but frequency information, the decoding unit 103 does not need to perform full decoding. For example, when the content analysis unit 104 analyzes AudioES, if the content analysis unit 104 calculates frequency information such as spectrum and MFCC, the decoding unit 103 does not perform full decoding to generate PCM. Decoding up to MDCT can reduce the process of converting MDCT to PCM and the process of converting PCM to frequency information.

  As a result, for example, when content is sequentially analyzed during recording of one-segment broadcasting and the analysis result and content are linked almost simultaneously with the end of recording, the user does not watch the program during recording, so the information is fully decoded. Without decoding, it is possible to greatly reduce the processing by decoding AudioES up to MDCT.

  The segment unit feature quantity extraction unit 104b calculates a feature quantity for each frame from the input decoded data, and uses the feature quantity for each frame to calculate a feature quantity for each segment composed of a plurality of frames. This value is discretized based on the reference value.

  Here, an example is shown in which the audio ES is separated and extracted from the received one-segment broadcasting by the Demux unit 101, and the segment classification unit 105 classifies the Audio ES into a section in which music is flowing and a section in which music is not flowing. The feature amount extraction processing in the feature amount extraction unit 104b will be described.

  In the case of a stereo sound source, the difference between L and R tends to be large in a section where music is flowing, and the difference between L and R tends to be small in a section where music is not flowing. Therefore, the segment unit feature quantity extraction unit 104b extracts the magnitude of the difference between L and R in segment units, discretizes the segment by the segment unit feature quantity (in this case, binarization), and the section classification unit 105 Classify into music and non-music segments.

  FIG. 4 is an image diagram showing an example of decoded data processed by the segment unit feature amount extraction unit 104b. In FIG. 4, the feature amount of a segment composed of M frames is extracted. The segment unit feature quantity extraction unit 104b receives decode data from the analysis control unit 104a. This decode data is frame data decoded by thinning out at a thinning rate set by the analysis control unit 104a. FIG. 4 shows an example in which frames are decoded at the decoding unit 103 at a decimation rate of 1/2, and the decoded frames among the content frames are indicated by hatching. The decoded frame data is input to the segment unit feature extraction unit 104b.

  FIG. 5 is a flowchart showing the processing of the segment unit feature amount extraction unit 104b. The process of the segment unit feature quantity extraction unit 104b will be described with reference to FIGS.

First, the segment unit feature amount extraction unit 104b acquires data decoded by thinning out frames at the thinning rate set by the analysis control unit 104a (S101). The setting of the thinning rate in the analysis control unit 104a will be described in detail later. For example, FIG. 4 shows a frame that is decoded every other frame (1), frame (3),.

When such decoded data is acquired, the segment unit feature amount extraction unit 104b calculates a feature amount for each decoded frame (S102). In order to classify the music section and the non-music section, the difference between L and R is calculated. In FIG. 4, the feature quantity of the mth frame of the nth segment is expressed as x _n (m).

  Then, the feature amount of the segment unit is approximated using the decoded frame feature amount (S103). The approximate value of the segment-based feature value can be obtained, for example, as an average vector of the feature values of the decoded frames as shown in FIG.

  Then, it is determined whether or not the difference between the approximate value of the segment feature amount and the reference value for binarizing the segment is equal to or greater than a predetermined threshold (S104). The reference value and the predetermined threshold value for binarization are values input from the analysis control unit 104a to the segment unit feature quantity extraction unit 104b as the first analysis requirement.

  If the estimated value of the segment-wise feature value is significantly above or below the reference value, the feature value of the segment unit may be calculated using the feature values of all the frames in the same segment. There is a low possibility that the determination of binarization will be overturned. Therefore, according to the value calculated as the feature amount, the thinning rate, etc., it is statistically examined in advance how far away from the reference value there is a possibility that the determination of binarization will be covered, and the analysis control unit 104a manages as threshold information corresponding to each condition. When the analysis control unit 104a controls the segment unit feature amount extraction unit 104b, the threshold value and the reference value corresponding to the thinning rate, the reference value, and the like are output as the first analysis requirement.

  That is, in step S104, it is determined whether or not to perform binarization based on the approximate value of the feature value of the segment unit by comparing the approximate value of the feature value and the threshold value. Therefore, if it is determined Yes in step S104, the segment feature value is binarized based on the approximate value of the segment unit feature value (S105 to S107).

  If it is determined No in step S104, if the decoded data received from the analysis control unit 104a is not data obtained by decoding all the frames in the segment (No in S108), the analysis control unit 104a is informed. In response to this, a request to change the frame thinning rate is output (S109).

  The analysis control unit 104a that has acquired the frame decimation rate change request from the segment unit feature amount extraction unit 104b specifies the frame decimation rate to the decoding unit 103 and outputs a decode request. The frame thinning rate after the change is set to be lower than the thinning rate before the change, and all or a part of the frame that has not been decoded before is additionally decoded. As a result, the proportion of decoded frames in the frames constituting one segment increases, and the feature amount of the segment can be estimated more accurately.

  Then, the segment unit feature amount extraction unit 104b determines whether the difference between the approximate value of the segment unit feature amount and the reference value is equal to or greater than a predetermined threshold (Yes in S104) or the features of all frames included in one segment. Until the feature amount of the segment unit is calculated using the amount (No in S108), the feature amount is approximated by changing the thinning rate, and the process of determining whether the binarization can be performed based on the approximate value is repeated. .

  The binarization result of the feature quantity for each segment determined by the segment unit feature quantity extraction unit 104b according to such processing is output to the section classification unit 105 as the first feature quantity.

  As described above, when binarizing the feature value of the segment unit, not all the frames included in one segment are decoded from the beginning, but the feature value of the segment unit is determined by the decoded data obtained by thinning out the frames. Since there are cases where binarization is possible, the amount of processing required for decoding can be reduced. Also, if the approximate value of the segment-wise feature value is close to the binarization reference value, add all or some of the frames that were not decoded so that the segment-wise feature value can be calculated more accurately Therefore, it is possible to suppress degradation in accuracy of feature quantity calculation.

  In step S104, the predetermined threshold value can be set to 0. In this case, steps S108 and S109 are not executed, and the feature value is binarized using the approximate value of the feature value calculated for each segment. It is possible to further reduce the processing amount.

  It should be noted that the frame decimation rate of the decoded data used when approximating the segment unit feature amount may be set according to the segment feature amount for which the feature amount has already been calculated. For example, if the feature value of the previous segment unit is near the reference value, the current segment is likely to have the feature value of the segment unit near the reference value. If the feature amount of the previous segment unit is far from the reference value, the initial thinning rate may be set higher.

  In addition, for example, linear prediction or spline interpolation is performed using a plurality of past data, and the feature amount of the current segment is predicted. If the prediction result is close to the reference value, the decimation rate is set to a low value and greatly increased from the reference value. If they are far away, the thinning rate may be set higher.

  In the above-described example, the difference between L and R is calculated as a feature quantity to determine musicality, and the feature quantity is binarized. However, what value is calculated as the feature quantity? It depends on whether the section is classified. For example, the power value, spectrum, mel filter bank cepstrum coefficient (MFCC), LPC cepstrum, similarity to a pattern predetermined for each class to be classified may be used depending on the class to be classified. Further, the case of binarization has been described, but even with discretization such as ternarization and quaternarization, the processing amount can be reduced by using such a method.

  The time continuation feature amount extraction unit 104c calculates a feature amount for each frame from the input decoded data. When a predetermined number or more of frames satisfying a predetermined criterion continues, the frame is set as a specific section. Extract.

  In the following description of the time continuation feature amount extraction unit 104c, a case where a silent section is extracted from content is used as an example. When a silent section is extracted from the content, the power is calculated for each frame as a feature quantity, and a frame having a power less than a predetermined reference value is “h”, and a frame having a power greater than the predetermined reference value is “ It is binarized as l ″. When a frame of “h” continues for a predetermined number of continuations (explained here as 3), the section is extracted as a silent section “H”.

  The decode data is input from the analysis control unit 104a to the time continuation feature amount extraction unit 104c. FIG. 6 is an image diagram illustrating an example of decoded data processed by the time continuation feature amount extraction unit 104c. First, every third frame shown by hatching in FIG. 6 is decoded by the decoding unit 103 and input from the analysis control unit 104a to the time continuation feature amount extraction unit 104c. The frame thinning is determined based on the number of continuations of the characteristic amount “h” for determining the silent section “H”. Here, every three frames are decoded in order to determine a silent period “H” when “h” continues for three frames.

  FIG. 7 is a flowchart showing processing of the time continuation feature amount extraction unit 104c. The time continuation feature amount extraction unit 104c acquires the decoded data from the analysis control unit 104a (S201). This decoded data is a frame decoded every three as shown in FIG. The time continuation feature amount extraction unit 104c calculates a feature amount for the acquired decoded data, compares the calculated feature amount with a predetermined reference value, and binarizes “h” and “l”. (S202).

  There is a possibility that a frame determined as “h” is a frame constituting a silent section (a section in which a predetermined number or more of “h” frames continue). Therefore, a frame determined as “h” is extracted (S203), and it is determined whether there is a silent section including the frame. First, the analysis control unit 104a issues a request to the decoding unit 103 to decode the frame immediately before the frame extracted in step S203 (S204). When the time continuation feature amount extraction unit 104c acquires the data decoded by the decoding unit 103, the feature amount is calculated for each frame, and “h” is determined depending on whether the feature amount is equal to or greater than a predetermined threshold. And binarized into “l”. (S205). If this frame is “h” (Yes in S206), a decoding request for the previous frame is output (S207), and the feature amount of the decoded frame is binarized. The processes in steps S205 to S207 are repeated until a frame whose frame feature amount is determined to be “l” appears. As a result, it can be seen from which frame the frames for which the feature amount is determined to be “h” are continuous.

  In addition, a request is issued from the analysis control unit 104a to the decoding unit 103 to decode the frame immediately after the frame extracted in step S203 (S208), and the feature amount of the decoded frame is calculated. It is binarized into “h” and “l” depending on whether or not it is greater than or equal to a predetermined threshold value (S209). If this frame is “h” (Yes in S210), a decoding request for the next frame is issued. It outputs (S211) and binarizes the feature quantity of the decoded frame. The processes in steps S209 to S211 are repeated until a frame determined to be “l” appears. As a result, it can be understood to what frame a frame whose feature value is determined to be “h” continues.

  By using this information, it is possible to extract a section in which three or more frames determined to be “h” are continuous as a silent section. Therefore, information on whether or not each frame is a frame of a silent section. Is output as the second feature amount (S212).

  In this way, the frame that was first decoded at a predetermined interval and decoded is decoded in detail around the frame whose feature value is determined to be “h”, so the decoding process is reduced without degrading accuracy. can do.

(Transmission error detection filter unit 102)
Next, the transmission error detection filter unit 102 will be described. FIG. 8 is a block diagram showing detailed configurations of the transmission error detection filter unit 102 and the decoding unit 103. The transmission error detection filter unit 102 includes a header search unit 102a and a transmission error determination unit 102b.

  The header search unit 102a searches the ES header input from the Demux unit 101 and divides the ES into frame units. The ES divided into frame units is output as frame data to the transmission error determination unit 102b and is also stored in the frame data temporary storage unit 103a of the decoding unit 103.

  A time stamp such as a PTS (Presentation Time Stamp) and frame data are input to the transmission error determination unit 102b. Each time a time stamp is input, the transmission error determination unit 102b calculates the time difference between the time indicated by the previously input time stamp and the time indicated by the time stamp input this time as the first time length. The time length from the ES frame in which the previous time stamp is input to the ES frame in which the current time stamp is input is calculated as the second time length. If the first time length and the second time length calculated in this way are equal, it is determined that there is no transmission error, and if the first time length and the second time length are not equal, transmission is performed. Judge that there was an error.

  In this way, the transmission error determination unit 102b can reliably detect whether there is an error due to packet loss by using the time stamp, and in which time zone normal frame data exists, It is possible to accurately determine in which time zone the frame data that may contain the abnormality exists. The transmission error information obtained by the transmission error determination unit 102 b is output to the TS data temporary storage unit 106 and also output to the section classification unit 105.

(Section classification unit 105)
Next, the section classification unit 105 will be described. The section classification unit 105 uses either or both of the first feature value output from the segment unit feature value extraction unit 104b and the second feature value output from the time continuation feature value extraction unit 104c, or both. Get the desired classification interval.

  For example, when information indicating the difference between L and R in segment units is input from the segment unit feature amount extraction unit 104b, the segment classification unit 105 classifies the segment into a segment where music is flowing and a segment where music is not flowing. Can do. Further, for example, using the information indicating the silent section output from the time continuation feature amount extraction unit 104c, the section classification unit 105 determines that the silent section is the interval between the silent section and the silent section. A period between silent sections is determined as a CM section.

  For the section classification processing in the section classification unit 105, the transmission error determination result input from the transmission error determination unit 102b is used. The process of classifying sections using the transmission error determination result input from the transmission error determination unit 102b will be described with reference to FIG.

  Note that it is assumed that the segment classification unit 105 receives an approximate value of the difference between L and R in segment units from the segment unit feature quantity extraction unit 104b, and the segment classification unit 105 detects a music segment.

  The segment classification unit 105 can determine that the musicality is high based on the approximate value of the difference between L and R in the segment unit input from the segment unit feature quantity extraction unit 104b (a segment having a large difference between L and R). And discretized into segments (segments with a small difference between L and R) that can be determined to have low musicality.

  In FIG. 9, a segment determined to have high musicality is indicated as “H”, and a segment determined to have low musicality is indicated as “L”. Then, the section classification unit 105 will be described as determining that the music section starts when, for example, three H segments continue, and determining that the music section ends when, for example, three L segments follow the music section.

  The segment input to the section classification unit 105 includes a case where all the frames constituting the segment do not cause a transmission error and a case where a part or all of the frames constituting the segment cause a transmission error. . In FIG. 9, a segment including a frame in which a transmission error has occurred is indicated by “x”.

  If no transmission error has occurred in all frames, the segment unit feature extraction unit 104b may binarize into “H” and “L” based on the approximate value of the difference between L and R of the segment. it can. However, if a frame in which a transmission error has occurred is included, there is a possibility that the approximate value of the difference between L and R of the segment is not accurately obtained. Therefore, the information on the transmission error detected by the transmission error detection filter unit 102 is input to the section classification unit 105, and the segment including the frame in which the transmission error has occurred is binarized stochastically.

  As a method of binarizing stochastically, for example, the occurrence pattern of “H” and “L” of a predetermined number of consecutive segments (for example, five) is learned in advance, and the location where the transmission error occurs is determined. By comparing the appearance pattern of “H” and “L” of the centered segment with the learning result, it is possible to predict an error occurrence place stochastically. Specifically, for example, in the case where two segments before and after the transmission error center are the pattern “HH × HL”, and “HHHHL” is 8 cases as the previously learned H and L patterns, “ When there are two cases of “HHLHL”, the probability of appearing in the “HHHLHL” pattern is higher than the probability of appearing in “HHLHL”. Therefore, the segment including the frame in which the transmission error has occurred is determined as “H”.

  As another method of binarizing stochastically, if there are many “L” s in the two or three segments before and after the place where the error occurred, the segment is also determined to be “L” and the adjacent segment is set to “H”. If there are many "", the segment can be determined to be "H". In this way, by classifying sections using the stochastic binarized feature values, sections can be classified with high accuracy.

  In the above description, the case is described in which the sections are classified using the segment-unit feature quantities determined by the segment-unit feature quantity extraction unit 104b. However, even if the sections are classified using the feature quantities for each frame. good.

  In addition, it is not limited to the said embodiment, You may change suitably in the range which does not deviate from the summary of this invention.

1 is a block diagram showing a configuration of a mobile phone according to an embodiment of the present invention. The block diagram which shows the structure of the content processing part of the mobile telephone which concerns on embodiment of this invention. The block diagram of the content analysis part of the mobile telephone which concerns on embodiment of this invention. The processing image figure in the segment unit feature-value extraction part of the mobile telephone which concerns on embodiment of this invention. The flowchart which shows the process of the segment unit feature-value extraction part of the mobile telephone which concerns on embodiment of this invention. The image figure which showed the example of the decoding data processed with the time continuation feature-value extraction part of the mobile telephone which concerns on embodiment of this invention. The flowchart which shows the process of the time continuation feature-value extraction part of the mobile telephone which concerns on embodiment of this invention. The block diagram which shows the structure of the transmission error detection filter part and decoding part of the mobile telephone which concern on embodiment of this invention. The processing image figure of the area classification | category part of the mobile telephone which concerns on embodiment of this invention.

Explanation of symbols

1 cellular phone, 51 control unit, 52 antenna, 53 RF unit, 54 communication control unit, 55 voice input / output unit, 56 display unit, 57 storage unit, 57a multimedia content DB unit, 58 antenna, 59 RF unit, 60 broadcast Receiving unit, 100 content processing unit, 101 demux unit, 102 transmission error detection filter unit, 102a header search unit, 102b transmission error determination unit, 103 decoding unit, 103a frame data temporary storage unit, 103b decoding processing unit, 104 content analysis unit 104a Analysis control unit 104b Segment unit feature extraction unit 104c Time duration feature extraction unit 105 Segment classification unit 106 TS data temporary storage unit 107 Analysis result integration unit

Claims (5)

An information processing apparatus for classifying a predetermined section from encoded content,
Decoding means for decoding by decoding the frame included in the content at a predetermined thinning rate;
Frame feature amount calculating means for calculating a feature amount of the frame decoded by the decoding means;
When discriminating and calculating the feature amount of a segment consisting of a predetermined frame group, the feature amount of the segment unit is calculated using the feature amount of the frame calculated by the frame feature amount calculating unit, and the calculated segment unit amount of If the difference between the feature quantity and the reference value is equal to or greater than a predetermined value, the segment feature quantity is discretized using the segment unit feature quantity, and the difference between the calculated segment unit feature quantity and the reference value is a predetermined value. If the value is less than the value, the frame decimation rate is changed until the difference between the segment unit feature amount and the reference value is equal to or greater than a predetermined value or the segment unit feature amount is calculated based on the feature amount of all frames. Segment feature discretization means for discretizing the segment feature quantity using the segment unit feature quantity obtained by decoding by the decode means
An information processing apparatus comprising: section classification means for performing section classification of content using the discretized segment feature quantity acquired by the segment unit feature quantity discretization means. An information processing apparatus for classifying a predetermined section from encoded content,
Decoding means for decoding by decoding the frame included in the content at a predetermined thinning rate;
Frame feature amount calculating means for calculating a feature amount of the frame decoded by the decoding means;
Determination means for determining whether or not the frame feature value calculated by the frame feature value calculation means matches the condition of the predetermined section;
A front and rear frames of the frame that is determined to match the condition of the predetermined section by the determining means, with respect to awe frames not been decoded by said decoding means, sequentially, the decoding by the decoding means, said frame The calculation of the feature amount by the feature amount calculation unit and the determination of whether the feature amount by the determination unit matches the condition of the predetermined section are repeated until a frame that does not match the condition of the predetermined section appears. Thus , an information processing apparatus comprising section classification means for determining the continuation of frames that meet the condition of the predetermined section . A communication means for transmitting and receiving data;
A transmission error detecting means for detecting a frame that has been transmitted by a communication error and cannot be correctly received among the frames of content received by the communication means;
The information processing apparatus according to claim 1, wherein the section classification unit probabilistically estimates a feature amount of a segment including a frame in which a transmission error has occurred by the transmission error detection unit from previous and subsequent segments. A content analysis program for classifying a predetermined section from encoded content,
A decoding function of decoding by thinning out frames included in the content at a predetermined thinning rate,
A frame feature amount calculation function for calculating a feature amount of a frame decoded by the decoding function;
When discriminating and calculating the feature amount of a segment consisting of a predetermined frame group, the feature amount of the segment unit is calculated using the feature amount of the frame calculated by the frame feature amount calculation function, and the calculated segment unit amount is calculated. If the difference between the feature quantity and the reference value is equal to or greater than a predetermined value, the segment feature quantity is discretized using the segment unit feature quantity, and the difference between the calculated segment unit feature quantity and the reference value is a predetermined value. If the value is less than the value, the frame decimation rate is changed until the difference between the segment unit feature amount and the reference value is equal to or greater than a predetermined value or the segment unit feature amount is calculated based on the feature amount of all frames. A segment feature discretization function that discretizes the segment feature using the segment unit feature obtained by decoding using the decode function.
A content analysis program, comprising: a section classification function for performing section classification of content using the discretized segment feature quantity acquired by the segment unit feature quantity discretization function. A content analysis program for classifying a predetermined section from encoded content,
A decoding function for decoding by decoding the frames included in the content at a predetermined thinning rate;
A frame feature amount calculation function for calculating a feature amount of a frame decoded by the decoding function;
A determination function for determining whether or not the frame feature value calculated by the frame feature value calculation function matches the condition of the predetermined section;
A front and rear frames of the frame that is determined to match the condition of the predetermined section by the determining function for awe frames not been decoded by the decoding function, sequentially, the decoding by the decoding function, the frame The calculation of the feature amount by the feature amount calculation function and the determination of whether the feature amount by the determination function matches the condition of the predetermined section are repeated until a frame that does not match the condition of the predetermined section appears. And a section classification function for determining the continuation of frames that meet the predetermined section condition .

Images