TW201108005A - Video search method and apparatus using motion vectors - Google Patents

Abstract

A use of 3D motion vector map for a video story search is provided herein. In a method of a video story search, parsing processes are respectively performed on data stream of an input search video file and a plurality of video files dedicated to be searched, and then a plurality of motion vectors are obtained accordingly. A plurality of corresponding motor vector maps are generated in time domain. A correlation result is generated according to the motion vector map of the input search video file and the motion vector maps of the plurality of video files, and then a search result is obtained based on the correlation result.

Description

201108005

P52980146TW 33280twf.doc/I VI. Description of the invention: [Technical field of the invention] Method and its device, and special wheeling conditions for video search The invention relates to a video search for a video content search Search method and device for finding. [Prior Art]

- At present, the search technology on the Internet is currently based on text-based technologies, such as Go〇gle, Yah〇〇, the nameless station in the country, and so on. Main = However, each search engine wants to break through the limit of text f ^, for example, hit, the search pin field is Simplified Chinese, and even the inner valley of other languages, but after all, shopping is limited by text keywords. For example, if you want to search for relevant media (4), including the sound age (4) case 1 2 because there is not enough text content to be used as a basis for search, or countries to use different translation names, resulting in Find out the right or more relevant information. Google search engine launched a photo-based photo service in April 2009. It is the world's first service to find content with photo content (c〇ntent). For example, referring to FIG. 1A, when the "aPPle" keyword 11 is input in the input box 120, related information related to "apple" appears, but this includes all images in the shape of Apple, and the trademark "Apple®". Related products mobile phone "iPh〇ne,,. At this time, if you further select, you can eliminate a lot of unsuitable data. For example, Figure 1B, after the user selects the image related to Apple's shape, the search engine further displays this Class 201108005

P52980146TW 33280twf.doc/I Fruit (Apple) related images. For example, in FIG. 1C, the user selects a mobile phone related to the trademark Apple®, and after the image, displays other images related to the product, and more accurately finds the photo that the user wants to search. However, this technology is obviously Use Image content to search for images, but only for the search of related photo files. There is no way to search for multimedia files. To break through this limitation, in Motion Picture Professionals (Motion Picture Expert Group 'hereinafter referred to as mpEG's technical agreement MPEP-7' proposes a standard with supplementary information for content, especially for multimedia digital content. In this MpEp_7 standard, relevant content descriptions can be provided for multimedia ( Multhnedia

Content

Description), and can be independent of other MpEp standards, and this digital description can even be attached to an analog movie file. As shown in Figure 2, 'for a parent's audio-visual content (AU (ji〇-visuai Content, as shown in the figure "AV Content") can be given a corresponding content description (c〇ntent Description) 'This content description is mainly proposed The relevant feature value of this audiovisual content, and the way of slotting is as shown in the figure: AV+Descript+AV+Desript+AV+Desript+... This AV stands for Audio-visual Content. Desnpt stands for the corresponding Content Description. However, such a structure is too complicated, and all multimedia files must be rearranged, which is not suitable for existing files and architectures. In addition, although it can be angered by the feature value' Find similar multi-201108005 by searching with similar keywords.

P52980146TW 33280twf.d〇c/I Media Archive 'But it is impossible to escape the gap between different languages by text search.

In addition, as the combination of the Internet and TV becomes more and more popular, video search on TV will inevitably encounter the problem of keyword input. The average person who watches TV is holding a remote control. The size and function of the remote control cannot replace the keyboard as a text input device. Therefore, using the remote control to control video search on the network TV is one of the future applications. The problem is where. SUMMARY OF THE INVENTION In one embodiment, a video search method is provided, including parsing a bit stream of an input search video file and a plurality of video files to be retrieved, and extracting a corresponding plurality of motion vectors. A plurality of corresponding motion vector maps are established on the time axis based on the motion vector. And according to the moving vector map of the search video file and the motion vector map of the video file, a Correlation degree is obtained, thereby obtaining a video search result according to the correlation degree. In a consistent example, a video search device is proposed, including a stream analysis, a 3D motion vector map generator and a 3D motion vector map. H ° This stream analysis H is used to locate bits of a video file. The stream is parsed (Parsing), and the motion vector (guess) data of the bit stream order is extracted according to the ratio of (5). The 3D motion vector map generator uses the motion vector to construct a 3D-MV Map with time_data. The ❼3D motion vector map comparator is configured to perform operations according to the 3-D motion vector map, and output a video slot according to the operation result.

The degree of correlation (COTelatkm) of the P52980146TW 33280twf.doc/I case, in order to obtain the results of the video search based on the degree of association. In an embodiment, a video playback device is provided having a host and control unit. The host has a video search device, and the video device includes a stream parser, a 3D motion vector map generator, and a 3D motion vector map comparison. This controller has - function control. The device, wherein the function control jianlong selects one of the video data played by the host as a search video file of the video search device. The above features and advantages of the present invention will become more apparent from the description of the appended claims. [Embodiment] 'This, the implementation example proposes a new video search technology, which breaks through the goal of the text-based search technology, which is based on the video content as a technology for searching for k-pieces. Search results. In one of the embodiments of the present invention, one of the video slots is selected as a Query Message, which is selected by the user to select any time-length subtraction or automatically via the user interface. The method of selecting the towel-segment gj or the specific (4) video age can be used in this example. In the embodiment, the user selection method can be built in control, and the remote control n-film of the video playing, such as a television or a DVD player, or the use of the built-in touch display or the screen. The user interface can be stopped in a convenient and simple manner by the user - the correction of the end time is the application of the present invention. ’,, 201108005

P52980146Γ W 33280twf. doc/I For the choice of film format as a shot, even the size of the face, due to its name or a movie, the plot is the same, J is different, but if it is the same as Vector, MV The distribution will be the same or similar, and its dynamic vector (M〇ti〇n will be introduced later. Therefore, as long as the index is shown in Figure 7 or 7B, the search index can be found. To establish a video, for example, a video file as a search condition = a similar plot. His video slot case can be used to retrieve all of its video files from the 袼 欲 。 。 。 ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ The f of the same format, so 'the corresponding video slot of the same timeline section in the fortune to be retrieved: it is difficult to find the video with the secret ^. In the conversion, his video building can be present. District;:=main==^main5 library, or cloud database, etc. The two-person host of this format conversion, the system server of the local area network, the host of the search engine or the cloud computing system Yes, in this example, convert to the same The purpose of the formula is to obtain the video slot for the = condition, and all other video to be retrieved, the Motion Vector (Mv), which is 5, for all compressed video files. The frame takes its motion vector (MV)' to establish a retrieval index. In one embodiment, a stream parser can be used to parse the data bit stream of all compressed video files (Parsing). And take out its motion vector (MV). For the motion vectors of different resolutions, in this embodiment, a statistical method for the motion vector (MV) is proposed, for example, by image group (Gr〇up 〇f 201108005

P52980146TW 33280twf.doc/I

Piet recommends 'GOP' as the basic unit, for all its blocks (μ_Block, MB) 'the size of the cage block to which the motion vector belongs (BlGek s(4), for example, a threshold (Thresh〇kJ) Decide which mobile vector (MV) to use to create the search index. For different video files, although the name or video format, even the size and quality of the face are different, if the same film, the plot is the same 'So its dynamic vector (MV) distribution will be the same or similar, as shown in Figure 7A or 7B, that is, the frame does not change much on the time axis. Therefore, it can be selectively scaled (for example 1:2) , 1:4, 1:N, N is an integer) Take the motion vector (MV) value of the corresponding frame (Frame). Therefore, the established -D motion vector map can achieve a certain degree of precision.

Private % only one of her target cases, according to the 3_D search: rr=. And the root of the case (also known as the search mark _ (four) mobile vector map check ^ ^ box (10) - Ling Gang's can value' and all the desire to carry out (Mic, L Lin case 31 mobile vector map corresponding to the micro-square, should The value of _ is the difference operation, the distribution of the progress of the situation, for the loss, == fruit (four) meaning ((10) know the process of the search for the face, ❺ in the step. Step 3g 'start the video to find the video operation '20 'Select one of them - segment video as a search W (Query 201108005

P52980146TW 33280twf.doc/I

Message). This selection method selects the video recorder of the length of time, whether it is played through the user's body, the remote control or other means of playing in the soft video -_(^Retrieving information during playback), or via the user The interface, after making the film as a function button, automatically selects 1 towel one, · ', one of which is other can be selected: the video in the range can be used in this example. . The wood 5 method, in the second step, as the step coffee, is thus, for example, the second case of the building as a search condition, optionally via format conversion =:= conditional reduction step county is the same as the money silk style, Kneeling does not need to be done. Since the addition of the embodiment of the present embodiment is changed to ===; =:, the video standard can be the presence of the network::I; or the cloud database and so on. The host of the m engine can be parsed (parsing) with all the video _ 'usually iil^rsram that has been compressed and selected, and the time axis can be selected; The purpose is to adjust the exchange rate flexibly. This video will cover all video files (including as a search [S 1 9 201108005]

P52980146TW 33280twf.doc/I conditional subtraction) is converted to a secret _ meaning, which is the motion vector of the frame within all the views. That is to say, for all (four) video broadcast = take out its motion vector, according to which the retrieval index is established. For the motion vectors of different resolutions, in the present embodiment, the change is performed by using up-sampling or down-sampling (D〇.Sampling). For example, 'general video slot slaughter=, many frames (four) (four) that are consecutively arranged according to the time_order are composed of a number of micro-blocks (Mien) BlGek, MB) and each micro-block MB is for example Is a 16xl6 riding position, and for each block MB has a motion vector, there may be one, or may have μ MB can be _ into 16 4 * 4 sub-blocks), so different formats of video light - MB may There are 丨~16 MV values in f after the _ rain difference calculation will cause the corresponding operation; therefore, in order to ^ ΐ degree, the mobile vector of each micro-block must be exempted from In the example, if the η moving vectors are to be integerd, the value of the n moving vectors can be calculated as an average. If you want to convert only one motion vector to have n movements, you can convert 1 motion vector to read the same value of movement - 徊ί outside' for how the mosquito micro-block has fun The vector is a case where 4 is n turns-- (can be done by statistical method. Face group έ fMPEP^1fl 编魏^, usually defines the book Ξ=(^ΡΓ^Γ00Ρ), example ^, Smart tragic shirt like h 'to achieve a better effect, will define this 201108005

P52980146TW 33280twf.doc/I Picture Group (GOP), you can obtain random access actions in the image data. For example, in the MPEP-4 agreement, it includes nine pictures (one picture, two P-pictures predicted by the shovel) B face with six bidirectional predictions). Therefore, if it is desired to know the number of mobile vectors to be used, in an example, a face group (GOP) may be a basic unit, and a block size (Block Size) to which the motion vector of the micro block MB belongs may be used. The ratio is taken, for example, by a threshold (Threshold) to determine the index of the search using that number of motion vectors. .

Then, referring to step 350', according to the 3D motion vector map (3D-MV Map) constructed by the motion vector (MV), the correlation value can be obtained after the operation, and the related video files can be found according to the correlation value arrangement. And = display the results according to the degree of relevance. After selecting the video building as the search condition, due to its name or film format, even the size and quality of the face are different, but if it is the same film, the plot is the same, so its dynamic vector (MV) The distribution will be the same or similar, as shown in Figure 7A or 7B. Therefore, as long as the search index (SearchIndex) is created for the selected rhythm project, a movie with similar plots can be found. In one of the implementation examples, for the manner in which the search index is created, a moving vector map (2D: MV Map) can be obtained based on the obtained movement (MV) 'stored in the operation matrix'. According to the selected interval of the subtraction case, if you are 30 seconds or one minute, and according to the continuous display of the frame (Ram produces 21 different motion vector maps, this considers the relationship of the time axis to create a 3-call) The movement of the D-moving vector map (3D_MV Map) is quite large due to the number of frames in a time interval, for example, 3 seconds in one second [S3 11 201108005

P52980146TW 33280twf.doc/I ^ Go to the face frame in t minutes. Therefore, if the corresponding shift U 'will require a lot of calculations, it may delay the processing for different video ages, depending on the name or shadow # format and quality, but if the same - The movie, 疋 the same, so its dynamic vector (MV) distribution will be the same or similar, as shown in ^ A or 7B. That is to say, the change in the face on the time axis is not large. It is possible to select (4) to - (for example, 1:2, 1:4, 1:N, N is the whole movement = the movement vector (Mv) value of the corresponding frame (Frame). Therefore, the established 3-D Moving a vector map can achieve a certain degree of precision.

In the meantime, one of the embodiments of the present invention lays down according to the 3-7 map, and finds the relevant video job. And the results of the search are displayed according to the degree of relevance. In an embodiment, the MV values of all the micro blocks (MBs) in the 3D moving vector map of the selected video 4# case (that is, the search target) can be compared with all the visual cases of the image to be retrieved. The MV value of the corresponding micro-block (Micro Block, MB) in the 3_D mobile vector map is subjected to difference calculation, and then according to the operation

The difference distribution is compared and the degree of correlation is obtained according to the result of the comparison as a basis for displaying the results. For example, in one embodiment, according to two video files for comparison, the motion vector value of the Nth frame, and the Nth frame of the other video file, the motion vector value is performed, for example, R〇〇t_Mean Square (RMS) = calculation or calculation of the absolute value of the difference, and obtain the difference (Distance), according to the difference of the two differences as the degree of association (C〇rreiati〇n) The result is displayed and the results obtained are displayed. 12 201108005

P52980146TW 33280twf.doc/I In the new video search method proposed by the present invention, in another example, please refer to the flowchart of FIG. 4, which is used to explain in detail how to construct a content as a search condition technology, and to find a video. Video search = first, the steps are complete, start to call for video to find the fall. Select one-segment video slot case as search information (c^y body broadcast) m2" select mode is not 4 疋 money users use video to play soft = case, or disagreement - the amount of the interface, in the time m, automatically select One of the fixed or specific period of the video slot 22 can reach the scope of the secret-time-dependent case, which is within the scope of the spine. After selecting the video file as the search condition, as in step 430, 3 name or video format 'even the size and quality of the face are different, so 'for example, the video credit check as the search condition = video_, can be selectively converted to have phase = format via format conversion Vision, but this step is not required if the video age format is the same. Since this embodiment adds the feature of the Time Domain, L can be used for all the viewings of the desired search. Find the video track corresponding to the same time axis 'and then' through the operation to convert to the video symbol $ with the same format. This same format can be pre-scheduled in the format, for example, all use in this method Format, the system operators or the pre-format piece goods can helmet, to be able to optimize the efficiency of the search results on the basis of the main considerations. '... 13 [s 201108005

P52980146TW 33280twf.doc/I Here are 3 examples + all that are to be checked in the database 432, for example, the host database that can be saved by the b check, or the cloud; ^; road == the host or the cloud of the engine The computing system can do it all. Search + in step 440, for all or part of the compressed data, the bit hit, ,, Mutong hanging as already and can be parsed from the string *) (ParSin§), and selectivity The ratio of the grading of the corresponding frame (for example, 1:2, 1.4 or (9), Bean II) is the sampling rate of the container. In this embodiment, the video file as a search condition is converted into the same format, and the mobile vector of all the video slots is used. That is to say, all compressed video jobs take out their motion vectors, based on which the retrieval line 51 ° 7 is established and the resolution for different retrieval conditions, that is, the movement direction of different resolutions, in this embodiment, Change it by up-sampling (Up-Samplmg) or down-sampling (D__Sampling). For example, a general video file is composed of a plurality of consecutive frames, and each frame is encoded by a plurality of micro-blocks, and each micro-block MB is, for example, 16x16. For each micro-block MB, there is a possibility that there may be one or six motion vectors, and there are different numbers of motion vectors for different formats. However, in order to unify the degree of resolution, the number of motion vectors that each micro-block MB has must be adjusted to be consistent. In one embodiment, if n mobile vectors are adjusted to one, 201108005

The value in the value is an average. In addition, if there is only one Ω. Vector, the vector, for example:

^ 1 ^v^Jiuup υι rictures > GOP) ^ Example? When the MPEP-4 protocol processes continuous motion pictures, in order to achieve a better compression effect, the surface group (GQP) will be defined, and random access actions in the image data can be obtained. For example, the ΜίΜ agreement includes nine pictures. (One I face, two forward predicted faces and six bidirectionally predicted faces). Therefore, if it is desired to know the number of mobile vectors to be used, in one example, the face group (GOP) can be the basic unit, and the block size to which the motion vector of the micro block MB belongs (B 1〇( ± size) The ratio, for example, Thresh〇ld, is used to determine the number of motion vectors used to establish the search index. For example, 'in the statistical face group (GOP), there is a block below. Block Size and its ratio: BS=16xl6 accounted for 5〇〇/0 BS=16x8 accounted for 15〇/0 BS=8x8 accounted for 25〇/〇BS=8x4 accounted for 3% 15 201108005

P52980146TW 33280twf.doc/I BS=4x4 accounted for 7%. In this example, the threshold (Thresh〇ld) is set at least 5〇%, so the block size is 16x16, which is 50%, so it is eligible. Then, the block size BS=16xl6 is selected, and the movement of the bit stream (Bit_m) is used as the size of the moving vector map resolution (1)叩Resolution. That is to say, regardless of the number of Mvs in the original MB, according to Na = call = ^ = secretion > gauge 4 = .__. = ^ way, are adjusted to one. However, if there is no threshold value of the ratio of the health block size, then the block size of a certain larger size, for example, the motion vector in the Bs=shirt, can be changed as the magnitude of the resolution of the moving vector map. Then, the adjustment is performed; that is, if there are 9 Mv values in the MB, the MV in the checker is also adjusted to 9 and the block size is also the same. _ ί Referring to step 45G, a 3D motion vector map (3D_MV Map) having a time axis > material is constructed according to the motion vector (MV), and the associated value can be obtained through the work, and the related view can be found according to the correlation value. Plant the case and display the results according to the degree of relevance. After selecting the video building as the search condition, since the video format 'even the size and quality of the picture is different, but ^ = the movie's plot is the same, its dynamic vector (MV) distribution will be similar or similar As shown in Fig. 7A or 7B. Therefore, as long as you create a search index (Sea]xh Index for the selected Shishi =^, you can find a similar story. In the implementation example, the material created for ^= (Search Index) can be based on the obtained movement = 201108005

P52980146TW 33280twf.doc/I Figure for example: (2,3) (2,5,2) (2,4) ~〇J) (4,3) T4J)^ (6,4) '(5,2) 5) According to the time interval of the selected video file, for example, % I, or - minute, and according to the continuously displayed frame, different ^ ^ vector maps are generated, and considering the relationship of the time axis, a = Moving vector map (3D-MV Map) data. However, due to the considerable number of faces in the second _, for example, 30 seconds in one second, and 1800 pictures in 8222. Therefore, if the corresponding motion vector is filled in, it requires a lot of calculations, which may delay the processing time. 〃 : The same field of view of the 曰 料 , , 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂 杂Shown in 7B. That is, the change of the face on the time axis ^= j This can selectively obtain the motion vector_value of the corresponding frame by a proportional (for example, 1:2, 1.4, (9) integer). Therefore, the 3-D moving vector map can achieve a certain degree of precision. The established data base and the application example are, as in step 460, based on the 3-D movement to determine the I map. And according to the degree of correlation, the search in the one:::, the selected video will be the target of the search) [S ] 17 201108005

P52980146TW 33280twf.doc/I All the micro-squares (10)(10) in the vector map are the same as the micro-squares (10) (10) corresponding to the .I. j. nv/T climbing diagram. (4) (4) The difference is guessed and then calculated according to the difference. The distribution is compared, and the degree of correlation (-) is obtained according to the results of the comparison, as the basis for displaying the results. For example, in one embodiment, according to two video games for comparison, one of the video frames, the Nth frame of the case (the Ffame) movement vector difference and another The difference between the _) of the fresh N box and the movement vector of the first frame of the frame is calculated by the root-Mean-Square (RMS) or the absolute value of the difference. Calculate 'and get the difference (Distance), turn the distribution of these differences as a result of the degree of Correlation, and display the results. For example, the difference is the value underneath: D = ^frame{n)MV ~frame{n - 1)Λ#"]2 is the calculation method using the root mean square, or:

D = \frame{n)MV - frame(n - \)MV According to the calculated difference (Distance), the difference distribution of the different motion vectors' can be correlated with the search target and sorted according to The results can be obtained, and the most relevant, for example, the top ten or the top twenty video files can be found and the results displayed in response to the user. 201108005

P52980146TW 33280twf.doc/I This method proposes a new video search method that can be applied to software. In other implementations, you can also use the design of the gift, such as matching the hardware of the part, and the material and operation of the soft rainbow. You can also use the method of (10) to perform the job search on the network. In addition, the new video search method proposed in the present embodiment may be placed on a personal host or a local area network, and the other part may be a system that can be placed on the end, such as a search engine host or a cloud design.

The new video search method proposed in this implementation can adopt the method of roughly selecting (c〇arse Seiecti〇n) in the manner of FIG. 1A to FIG. 1C, first finding a plurality of related video files, and responding to the use. Then, a more relevant video is found in a way of Fine Selection (closer to improve the comparison condition), which is also one of the application modes of the embodiment. For hardware applications, one of the various embodiments may be illustrated with reference to the circuit blocks of FIG. 5A, but is not limited thereto. In this circuit application, including stream parser (stream)

Parser) 530 ' 3-D motion vector (3D-MV) map generator 540 and 3-D motion vector (3D-MV) map comparator 550. In the architecture of the stream parser 530, bitstreams of a plurality of already compressed video files can be parsed and their motion vector (MV) data can be fetched. For the resolution under different retrieval conditions, that is, the motion vectors of different resolutions, the stream parser 530 can be changed by using up-sampling or down-sampling. In addition, in order to unify the search index (SearchIndex) resolution, 19 [£1 201108005

P52980146TW 33280twf.doc/I string, :L parser 53G can adjust the number of movements of each micro-block MB to -" For example, if you want to adjust the η-na motion vector to 里, use, for example, the averaging method, The value of the η homing vector is done as a flat ^ calculation. However, if only one moving vector is to be converted to have a moving direction =, one moving vector is transferred to the same value of the moving two additional 'stream parser 530 can be used for how to select the micro block ΜΒ The number of moving vectors is done by the secret method. For example, in the group (coffee) as a basic unit, the ratio of the block size of the mobile device of the micro-block is determined by, for example, a threshold (Threshold). The motion vector is used to establish a retrieval index.

The 3D-MV map generates n 54G, and the simple axis vector constructs a 3D motion vector map with time axis data (3D_MV applied. This MV map generates benefit 540 according to the obtained motion vector (MV), and stores it into an operation matrix. In the middle, get a 2_D moving vector map (2d_mv Map), and generate different 2_D moving vector maps according to the time interval of the selected video file and the continuously displayed frame, and then add the parameters of the & Then, a data called a three-seven motion vector map can be created. Since there are quite a lot of faces in a time interval, the 3D_Mv map generator 540 can select a certain ratio (for example, & 1:4, 1: n, N is an integer) to obtain the motion vector (mv) value of the corresponding prame. Therefore, the established 3-D motion vector map can achieve the set accuracy. In the -κ & Two stages can be used, including a rough selection of 20 201108005

P52980146TW 3328Otwf. doc/I (Coarse Selection) and the way to find the most relevant video in a precise way. The 3D-MV map comparator 550 determines the relevant video file based on the ίο moving map. And the results of the search are displayed according to the degree of relevance. The 3D-MV map comparator 55 is connected to the network 552 or the database 554 to read the 3D-MV map. The 3D-MV map comparator 55 can take the difference between the Nth frame of the video file and the disco vector of the first picture (F job e) according to the two video files being compared. And another video file, the Nth frame (Frame) and the N-1th frame (Frame) movement vector difference 'both, for example, the root mean square (R〇〇t-Mean_Square, RMS) calculation is The calculation of the difference absolute value 'and the record __), according to this = difference 5 cloth: J correlation (Correlati〇n) degree results, and the results obtained are not on the display 560. (d) In the circuit of the example, it also includes the video format ΐ = ancient, the visual 'for the searched video file 51G ' and the corresponding video frame selection for the same time axis segment. Calculate the video building in the same format. This; r standard, but if ===== will directly be used as a search for video posts 51 rumors from the broadcast case; in the implementation example '* -' this choice is made by making 201108005 P52980146TW 33280twf.d〇c /l 二 机 The body can use the energy control device 'where the function control device is used to sell the video data in the segment - TV, and control; in the embodiment, the host can be a video camera. In the other! ^3 wire connector 'wireless control of the electric controller is wireless or has _ mouse. The stomach host, for example, 'clearly refer to Figure 5Β, can choose to cover an ip. sy in the control of the video playback of the remote control crying can be built-in playback, when the film is in the TV 57〇T2 end, to At the beginning of the second time, the video between m was used as the retrieval information. In another implementation example, the user interface of the user, such as a mouse or a user interface on the touch screen, can be used to conveniently and simply stop the video at one end as a search for the user. There is a movement; mouth (10) (1) 疋 in the video job t, according to a series of consecutive frames, and a schematic diagram of its motion vector. And these frames 620, 622, 624, 626, and 628 can be used by LV you to reach the county ((10)). Figure 6C shows the different block sizes (Block Size) to which the mobile vector of the micro-blocks belongs. For example, 630 shows that the block size is ΐ6χΐ6, 22 201108005

P52980146TW 33280twf.doc/I has - a motion vector. 631 shows two moving vectors. 632 has a block size of , and 633 shows a block size of δ χ 8 different zonal size vectors. 634 shows that the block size is fixed ^* has four moving rnr and has a moving direction. 6 5 ^ 636 is the block size of 8 different directions in different directions, and eight moving vectors. ❿637 shows that the block size is 4x4, and: has a motion vector. sweet

Please refer to FIG. 7A and FIG. 7Β, respectively, for the same movie (the film name is (4)' but with different linings, such as high resolution (m(four) efmiti〇n, HD) movie segment 71G and lower resolution general image format (Common In the lmage format, ciF format, the movie clip 73〇 (the same time • k 4 minutes and 18 seconds). According to the plot of the movie (that is, the change of the time axis), there are 720, 722, 724, 726, 72δ as shown in the figure. Frame, and the frame of 740, 742, 744, 746, 748. However, as can be seen from this content, for different video slots, although the name or video format, even the size and quality of the face are different, but if In the same film, the plot is the same, that is, there is little change in the timeline. In the specific verification example, please refer to Figure 8Α to find the movie A (QCIF format) 810 and the movie A (CIF). Format) 820 is associated with another unrelated movie B (CIF format) 830. This common image format (CIF) is 352x288 pixels, and QCIF is called a quarter-universal image format (Quarter). Common Image Format, QCIF), general size It is 176x144 pixels. After using the method proposed in this implementation to calculate the video on the network, the difference is shown in the figure.

23 201108005

P52980146TW 33280twf.doc/I 8B shows that the reference numeral 812 is the difference of the movie A (QCIF), the reference numeral 822 is the difference of the movie A (CIF), and the reference numeral 832 is the difference of the movie B (CIF). From this result, it can be understood that the difference distribution of the film A (QCIF) is very similar to the difference of the film a (CIF), but the difference between the first two and the film B (CIF) is very different, so the degree of association can be found. In addition, please refer to Figures 8C, 8D and 8E respectively, respectively, for the film a (QCIF format), the film A (CIF format) and the film b (CIF format) under the condition that 23=8, the obtained 3D- MV map distribution, it can be more clearly seen that the 3D-MV map distribution of film A (QCIF) is very similar to the 3D-MV map distribution of film A (CIF), but the former two and the 3D-MV of the film b (cif) The map distribution is very different. It can be seen from the above empirical results that the new video search method proposed in the present embodiment can obtain the value of the degree of association with the search target according to the calculated difference distribution, and can obtain the result according to the sorting, and can obtain the result. The result shows a response to the user. Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and any of the technical fields of the present invention can be modified without departing from the spirit and scope of the present invention. _, therefore, the protection of the present invention is subject to the definition of the towel. BRIEF DESCRIPTION OF THE DRAWINGS Figs. 1A to 1C are diagrams showing a conventional search method for finding an image by name. Figure 2 shows the audiovisual content in the MPEP_7 standard (av & curry 24 201108005

P52980146TW 33280twf.doc/I Schematic diagram of association with the Content Description. FIG. 3 is a flow chart of a method for searching for a break on the network according to an embodiment of the present invention. 4 is a flow chart showing a method for performing video search on a network according to another embodiment of the present invention. Fig. 5A is a block diagram showing the circuit implementation of an embodiment of the present invention. </ RTI> FIG. 5B is a schematic diagram showing an application method for selecting a retrieval movie segment in an embodiment of the present invention. Figure 6A is a schematic diagram showing the motion vector in a frame. Figure 0 is a schematic diagram showing a series of consecutive frames and their motion vectors in the video file according to the time axis. Fig. 6C is a diagram showing a different block size (Block Size) to which the motion vector of the micro-block MB belongs. 7A and 7B are schematic diagrams showing successive blocks of a segment and a display of different resolutions for the same movie, respectively. Figure 8A shows a schematic diagram of different movies, including the film's positive format, movie A (CIF format) and movie B (CIF format). The figure is a schematic diagram showing the difference values obtained by the three different films in Fig. 8A calculated by the method for video search on the network. 8C 8D and 8E are respectively a distribution map of the scam map obtained by the movie a (qCIF format) of FIG. 8A, such as the film (CIF format) and the scene multi-chip B (CIF format).

25 [S 201108005

P52980146TW 33280twf.doc/I [Key component symbol description] 530 : Stream parser 540 : 3-D motion vector (3D-MV) map generator 550 : 3-D motion vector (3D-MV) map Comparator 552: Network 554 = Library 560: Display. 610, 620, 622, 624, 626, and 628: Frames 710, 730: Movie Segments 720, 722, 724, 726, 728: Frames 740, 742, 744, 746, 748: Box 81〇: Movie A (QCIF format) 820: Movie A (CIF format) 830: Movie B (CIF format) 812: Difference distribution of movie a (QCIF) 822: Movie A (CIF) Difference distribution 832: difference distribution of film B (CIF), 26

Claims (1)

201108005 P52980146TW 33280twf.doc/I VII. Patent application scope: 1. A video search method, comprising: selecting a video file having a length of time as a search visual file, and searching the video file and multiple video files to be searched. The file ^ ^ stream is parsed, and the corresponding multiple motion vectors are extracted; 卞, the bits sequentially create a plurality of motion vector maps on the time axis according to the motion vectors; and the movement vector of the video file according to the retrieval The map, together with the moving vector maps of the cases, is correlated (correlation) to the result of the video search based on the degree of correlation. x 2. For the video search method described in item 1 of the patent application scope, the search video file is selected by the user for the length of the video slot 3' as described in claim 1 of the patent search method. , Ι φ broadcast case is selected through a user interface link - length of time 4. If you apply for the job search method described in the full-time 丨 丨 item, ^ 索 讯 槽 案 些 欲 欲 欲 欲 欲 欲 欲 欲 欲 欲 欲 欲 欲烚=The video search method according to item 1 of the patent application scope, wherein the === case and the video buildings to be searched are converted into the video having the same format after the conversion is performed. The video search method according to item 5 of the phase range, wherein the case corresponds to the number of its movements Ο ^ ' 27 [SI 201108005 P52980146TW 33280twf.doc/I 7. The video search as described in item 5 of the application of the S Green patent scope The method, wherein the conversion is different for the search video file and the plurality of frames in the video file to be searched, having different numbers of the motion vectors, using up-sampling or down-sampling Change to -. 8. The video search method according to claim 7, wherein the conversion operation selects the frame to be adjusted to a certain number of the motion vector modes, and the pair is based on a face group (GOP). Unit, find the number of all the motion vectors of all the frames in the group, if the number exceeds a threshold, it is used as the basis for adjustment. 9. The video search method according to claim 1, wherein the video files to be searched are stored in a host of a local area network, a search engine host database, or a cloud database. 10. The video search method according to claim 2, wherein according to the shift, the vector establishes a multi-lobed machine vector map on the phase axis, and the search is performed on the time axis in a proportional manner. The video file and the frames in the video files to be retrieved are generated for the motion vectors of the selected frames. The video search method according to claim 10, wherein the ratio is the selection of the enzymes in the search video slot and the video files to be searched on the time axis. Is an integer, and Ν is greater than or equal to one. /, r is I2. For the video search method described in the scope of the patent scope, the basin = Ϊ map is filled in the selected frames - the operation matrix: the 2D-MV Map, In addition, the frame 28 201108005 P52980146TW 33280twf.doc/I 挑选 can be selected according to the time axis, then a 3D-moving vector map (3D-MV Map) can be established. 13 · The video as described in claim 1 a search method, wherein the mobile vector map according to the search video slot is associated with the motion vector map of the video broadcast, and the motion vector of each two adjacent frames in the motion vector map according to the search video file is compared. The operation difference value is compared with the operation difference of every two adjacent frame motion vectors in the moving vector map of the mother's video frame, and the correlation degree is obtained according to the comparison result. 14. The video search method according to claim 13, wherein the operation difference value of each two adjacent frame motion vectors is based on a root mean square operation or a difference absolute value operation. 15. A video search device comprising: a stream parser 'Parsing a bit stream of a plurality of video files' and extracting a plurality of motion vector (MV) data; a 3D motion vector map a generator for constructing a 3D motion vector map (3D-MV Map) having time axis data according to the motion vectors; and a 3D motion vector map comparator for determining according to the 3_d motion vector maps, and Find out the degree of Correlation of the video files, so as to obtain the results of the video search according to the degree of association. 16. The video search device of claim 15 further comprising a video format converter for selectively searching for a video channel 29 201108005 P52980146TW 33280twf.d〇c/I and a plurality of The retrieved video file is converted to be parsed to the stream parser. (4) Round-up 17. In the video search device described in item π of the scope of the patent application, the video file is for the user to select a time-length view file: as described in claim 16 of the patent scope In the search device, the search video resource is a video file via the user-centeredness. _ _ _ 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. 19. The video search device described above, the video search for the search for the video service in the district _ ^ search engine host database, or the cloud database. Search for the video search device described in claim 16 of the patent scope, basin In the video file (4), the video file _ multi-side frame, ς 3 amount = motion vector, using the up-sampling or down-sampling method to change to be consistent. 22. As described in claim 21 The video search device, the flow analysis frame is converted to a certain number of the movements ,, and the 昼G group P is the basic unit, and the group is found. The number of all the moving vectors of the box is 'right exceeded: the number of thresholds is used as the basis for the adjustment. The video search device of the Chinese character 3D, such as the mobile phone, is based on the movements. The vector establishes a corresponding moving vector map on the time axis. The frames in the video files are selected at a certain ratio on the time 30 201108005 P52980146TW 33280 twf.doc/I axis and the motion vector map is generated for the selected motion vectors of the frames. The video search device of claim 23, wherein the certain ratio is selected on the time axis in the search video file and the video files to be searched for each of the Si pictures. A frame, wherein N is an integer, and N is greater than or equal to one. 25. The video search device of claim 23, wherein the 3D motion vector map generator fills the selected frames into a different A matrix is obtained to obtain a 2D-moving vector map (2D-MV Map). And, if the frames are selected according to the time axis, a 3D-moving vector map (3D-MVMap) can be created. The video search device according to Item 15, wherein the 3D motion vector map comparator compares each of the two phases in the motion vector map according to the search video file according to the motion vector map of the video files. The operation difference value of the frame movement vector is compared according to the distribution, and the degree of correlation is obtained according to the result of the comparison. 27. The video search device according to claim 26, wherein each two adjacent pictures The operation difference of the frame movement vector is based on the root mean square operation or the difference absolute value operation. _ 28. A video playback device having a host and a controller, wherein the host has the 15th item of the patent application scope The video search device has a function control device, wherein the function control device is configured to allow a user to select one of the video data played by the host as a search video file of the video search device. 31 IS] 201108005 P52980146TW 33280twf.doc/I (4) Please select the video reduction device described in item 28 of the scope, and wirelessly control the wireless controller to be a wireless remote controller for 30. The host is a video playback device as described in the "Computer ^^ item", and the controller is a mouse. 32