50 results about "Video copy detection" patented technology

The invention discloses a system and a method for parallel video copy detection. The method includes the steps of 1, selecting key-frames of a query video and a reference video by a parallel method and extracting MIFT features of the key-frames; 2, clustering extracted feature data of the reference video by a parallel hierarchical clustering method; 3, quantizing the features of the query video and the reference video by a quantitative method according to clustered results; 4, establishing indexes of quantized data of the reference video; and 5, retrieving by the parallel method, preliminarily searching in the indexes to obtain an alternative video by utilizing quantized data of the query video and then computing space consistency and time consistency to finally confirm a copy video. According to the system for parallel video copy detection, a parallel mechanism is adopted on the basis of fast retrieval of visual vocabulary bag model BOF (beginning of file), and accordingly detection efficiency of the system is improved greatly.

A two-step matching technique is embodied in a video-copy-detection algorithm that detects copies of video sequences. The two-step matching technique uses ordinal signatures of frame partitions and their differences from partition mean values. The algorithm of this invention is not only robust to intensity/color variations it can also effectively handle various format conversions, thereby providing robustness regardless of the video dynamics of the frame shots.

The invention discloses a video copying detection method based on robust hash. The advantages of spatial-temporal union information and a sequence characteristic in the aspect of reflecting video contents are used fully. The method comprises the following steps of: firstly, dividing a video frame into parts which are overlapped with one another in 50 percent and generating a time domain information representative image by using the video frame of each part; secondly, filling the time domain information representative images by using a hilbert curve to further calculate a gray scale three-meanvalue of each block on the hilbert curve, and generating a hash fingerprint by comparing the three-mean values; and finally, performing hash matching according to the obtained hash fingerprint and judging whether the video is a copied video. An experimental result shows that the robustness and the discrimination of an extracted characteristic are greatly improved and developed.

The invention belongs to the field of digital media and provides a video copy detection method based on compact video representation. The method comprises following steps: densely extracting keyframes of video library and inquiring video; extracting the image sparse features of the keyframes of the video library and the inquiring video; by means of the pooling mode, the image sparse features of the video library and the inquiring video are merged to form concise video features. The beneficial effects of the invention are: by means of the method, video information can be accurately described; feature data is effectively reduced; the retrieval is accelerated greatly; the method combines deep learning method and traditional method; on the premise of accurate matching, machine performance load is reduced and the problems in the prior art are solved.

The invention provides a video copy extraction method based on tensor decomposition, in particular to a video copy detection method based on multimodal features and tensor decomposition, which includes the following steps: (1) video preprocessing: video clips are standardized by spatial-temporal sampling; (2) video tensor modeling and tensor decomposition: the global, local and time domain features of the video are respectively extracted, tensor modeling is carried out, and the Tucker model is utilized to carry out tensor decomposition, so that a nuclear tensor and a low-order tensor are obtained; (3) video fingerprint matching: the nuclear tensor is utilized to carry out rough matching, and a video fingerprint is utilized to carry out fine matching in a coarse selection. Compared with the prior art, the method realizes the true complementary fusion of multimodal features of a video, not only overcomes the defect of poor robustness of video fingerprints constructed with single-mode features, but also realizes the temporal associated co-occurrence between a variety of modes of features, and increases the accuracy and efficiency of video copy extraction.

The invention discloses a video copy detection method based on the deep learning and graph theory, and aims to improve the speed and accuracy of image frame similarity matching. The technical scheme is to extract an image frame in a reference library video, extracting features of the image frame, and obtaining a feature matrix of the image frame in the reference library video; extracting an image frame in the video to be detected, extracting features of the image frame, and obtaining a feature matrix of the image frame in the video to be detected; performing similarity matching between the feature matrix of the image frame in the video to be detected and the feature matrix of the image frame in the reference library video by using an approximate nearest neighbor search algorithm, and obtaining a similarity matching list; and judging and locating video copy fragments by means of the graph theory according to the similarity matching list. The method of the invention can be used to cope with various video editing transformations and an arbitrary matching form between the video to be detected and the reference library video, solve the technical problem that the "noise" has a great influence on the matching effect and improve the speed and accuracy of the similarity matching of the image frames.

The invention provides a video copy detection method based on time-domain visual attention. The method comprises the following steps of: acquiring visual attention changes between different video frames according to a visual attention mechanism, and acquiring representation of time-domain attention degree; calculating a time-domain attention weight of a video frame in a video clip according to the time-domain attention degree to form a visual attention transfer image of the video clip; and finally, extracting video hash on the generated visual attention transfer image provided with time domain and space domain information. By the method, the time domain information of the video is fully considered, a video frame which makes the video content prominent is weighted, the time domain and space domain information is integrated through the extracted characteristics, and the video copy detection method has high robustness on time domain attacks.

The invention discloses a deep feature fusion video copy detection method based on an attention mechanism, and the method comprises the steps: (1) extracting a frame image from video data, and constructing an image pyramid through employing different scales; (2) taking the deep convolutional neural network model as a basic network, and adding an attention mechanism into a middle convolutional layer of the deep convolutional neural network model; (3) inputting the frame image and the image pyramid into a deep convolutional neural network model of an attention mechanism, and obtaining fusion features through splicing and fusion; (4) training a deep convolutional neural network model in a metric learning mode; and (5) obtaining source video data through similarity calculation by utilizing thetrained deep convolutional neural network model. According to the method, an attention mechanism and fusion of global features and local features are utilized, the problems that a traditional image processing method is low in efficiency and low in precision can be solved, and the problem that the global features cannot adapt to various complex changes can also be solved.

The invention discloses a video fingerprint extraction method based on slowly-changing visual features. The method comprises: For each image in a training set, an image of a random distortion versionis generated, spatial features are trained by using an original image and a distortion image, and parameters of a network are extracted; spatial and sequential normalization preprocessing is carried out on video data of the training set, wherein the data are processed into ones with a fixed dimension and a fixed frame number; a feature sequence of the network is extracted based on the spatial features and the extracted sequence is used as training data and an LSTM network is trained; and a network is extracted from the trained spatial features, the obtained features are in cascade connection with the LSTM network, and a video fingerprint is extracted. According to the method disclosed by the invention, the visual perception principle of the human being is simulated and thus the video fingerprint can be extracted. The method has the high robustness and high efficiency. The video fingerprint extraction method can be applied to fields of video copy detection and video retrieval and the like.

The invention provides a quadruple convolutional neural network video fingerprint algorithm. The algorithm comprises the steps of establishing a projection excitation network; constructing a quadrupleconvolutional neural network video fingerprint algorithm according to the projection excitation network; through video data selection, inputting the constructed quadruple video sequence into a quadruple convolutional neural network to carry out training and performance testing of the quadruple convolutional neural network. According to the invention, mapping from original video data to discrete binary codes can be realized in an end-to-end manner; simplified algorithm complexity, during training, tetrad loss and quantization error loss are used for jointly optimizing network parameters; on one hand, the intra-class variance is reduced and the inter-class variance is increased due to the loss of the tetrad; and on the other hand, the quantization error loss can reduce the loss of semanticsimilar information in a real value feature binarization process, the precision ratio and recall ratio of the method in the aspect of video copy detection are obviously improved, and the obtained video fingerprint can maintain relatively high robustness and uniqueness while meeting compactness.

The invention relates to the field of video classification and provides an end-to-end video copy detection method based on deep learning, which aims to solve problems in the video copy detection, suchas it is difficult to detect multiple copy segments in two pieces of videos, positions of copy video segments cannot be accurately positioned and the like. The specific embodiment of the method includes the following steps: performing shot segmentation on two video segments to be detected which are used for video copy detection, so as to select key frames; recognizing selected a plurality of keyframes by using a pre-built copy relationship recognizing model, and determining a copy relationship among the key frames; constructing a copy relationship matrix of all the key frames of two segmentsof the to-be-detected videos, according to the obtained copy relationship among the key frames; using the copy relationship matrix as an input of the pre-built positioning recognition model, and positioning fragments that contain a copy relationship in the two segments of the to-be-detected videos. The end-to-end video copy detection method and device based on deep learning can quickly and efficiently detect the video fragments that have the copy relationship in the two segments of the to-be-detected videos.

The invention provides a video copy detection method and device, a storage medium and an electronic device. The method comprises the following steps: extracting a first target key frame of a query video; extracting a first video feature of the first target key frame; acquiring a second video feature of a second target key frame of the reference video; and when it is determined that the query video is matched with the reference video according to the first video feature and the second video feature, extracting similar video clips in the query video and the reference video. The problems of low video copy detection efficiency and poor video copy detection accuracy in a video copy detection method in the related technology are solved, and the effects of improving the video copy detection efficiency and improving the video copy detection accuracy are further achieved.

The invention discloses a method and system for identifying a harmful video. The method comprises the following steps of when it is judged that the page elements of the webpage comprise the URL path of the video, identifying a user ID recorded in the page content of the webpage, acquiring a domain name contained in the URL or an IP address pointed by the URL according to a URL path of a video, andoutputting a first weight factor and a second weight factor based on the user ID, the IP address and a related query of the domain name; moreover, obtaining the video content of the film tail, carrying out the video copy detection on the video content of the film tail in a preset harmful video database, and outputting a third weight factor according to a monitoring result; and combining the firstweight factor, the second weight factor and the third weight factor to identify whether the video belongs to the harmful video. According to the method and the system, by combining a database made bythe big data, using the image processing means as few as possible and utilizing multiple modes, a scheme for identifying the harmful videos is provided.

The invention discloses a video copy detection method based on depth local features, which comprises the following steps: (1) extracting frame images from video data, and then constructing an image pyramid by using different scales; (2) constructing a deep convolutional neural network model for extracting a feature map from the input image pyramid and performing feature fusion on the feature map to obtain a fused feature map; (3) training the deep convolutional neural network model in a metric learning mode; (4) extracting a fusion feature map from an image pyramid by using the trained deep convolutional neural network model; (5) extracting key points from the fusion feature map by using maximum suppression, and extracting corresponding local features according to the key points; and (6) performing video copy detection according to the local features. According to the method, the extraction speed is higher, and the local feature representation is stronger, so that accurate detection can be carried out on various complex transformed copy videos and local features, and the method has the characteristic of high robustness.