73results about How to "Reduce information loss" patented technology

An apparatus for capturing a video image comprising a means for generating a digital video image, a means for classifying the digital video image into one or more regions of interest and a background image, and a means for encoding the digital video image, wherein the encoding is selected to provide at least one of; enhancement of the image clarity of the one or more ROI relative to the background image encoding, and decreasing the video quality of the background image relative to the one or more ROI. A feedback loop is formed by the means for classifying the digital video image using a previous video image to generate a new ROI and thus allow for tracking of targets as they move through the imager field-of-view.

A method for stabilizing a knowledge graph includes: generating a knowledge graph in which same entities in a semantic relation list between entities provided as an input are represented as a single node based on names and types of the entities; computing, on the knowledge graph, semantic similarities between all potential entity pairs of same entity types by comparing, for each potential entity pair, a type of relation associated with an entity in the entity pair and an opponent entity to the entity; and selecting, based on the semantic similarities, a representative entity from each of semantically similar entity pairs on the knowledge graph and integrating an opponent entity to the representative entity into the representative entity. The method further includes computing relation weighted values between the entities by using a graph analysis and statistic information, and adding the weighted values to the knowledge graph.

A complete User Datagram Protocol (CUDP) is disclosed that reduces packet loss. Channel frame error information is used with a packet level forward error correction (FEC) coding technique to accommodate wireless multimedia traffic. Each packet, as well as the channel frame error information, is forwarded to a given application. The CUDP protocol further assists the FEC decoding process by forwarding the locations of corrupted frames to the FEC decoder. Maximal Distance Separable (MDS) codes can be applied to a group of packets, to achieve additional robustness. An MDS decoder utilizes the frame error information to recognize the erasures within each packet. The error information can be represented as a set of LTU error indicators associated with each packet (for FEC decoders requiring an erasure indicator). The error indicators point to the starting and ending location of the erroneous data. The error information can also be represented as a reformatted packet (for FEC decoders Recognizing Erasures). The frame (LTU) error information from the lower layers is incorporated in the packet payload. An FEC encoder is also disclosed that encodes multimedia packets utilizing a packet-coding scheme, such as a Vertical Packet Coding (VPC) scheme or a Long Vertical Packet Coding (LVPC) scheme.

A complete User Datagram Protocol (CUDP) is disclosed that reduces packet loss. Channel frame error information is used with a packet level forward error correction (FEC) coding technique to accommodate wireless multimedia traffic. Each packet, as well as the channel frame error information, is forwarded to a given application. The CUDP protocol further assists the FEC decoding process by forwarding the locations of corrupted frames to the FEC decoder. Maximal Distance Separable (MDS) codes can be applied to a group of packets, to achieve additional robustness. An MDS decoder utilizes the frame error information to recognize the erasures within each packet. The error information can be represented as a set of LTU error indicators associated with each packet (for FEC decoders requiring an erasure indicator). The error indicators point to the starting and ending location of the erroneous data. The error information can also be represented as a reformatted packet (for FEC decoders Recognizing Erasures). The frame (LTU) error information from the lower layers is incorporated in the packet payload. An FEC encoder is also disclosed that encodes multimedia packets utilizing a packet-coding scheme, such as a Vertical Packet Coding (VPC) scheme or a Long Vertical Packet Coding (LVPC) scheme.

The present invention describes a road terrain detection system that comprises a method for classifying selected locations in the environment of a vehicle based on sensory input signals such as pixel values of a camera image. The method comprises a high level spatial feature generation for selected locations in the environment called base points. The spatial feature generation of the base points is based on a value-continuous confidence representation that captures visual and physical properties of the environment, generated by so called base classifiers operating on raw sensory data. Consequently, the road terrain detection incorporates both local properties of sensor data and their spatial relationship in a two-step feature extraction process.

Binning of predictor values used for generating a data mining model provides useful reduction in memory footprint and computation during the computationally dominant decision tree build phase, but reduces the information loss of the model and reduces the introduction of false information artifacts. A method of binning data in a database for data mining modeling in a database system, the data stored in a database table in the database system, the data mining modeling having selected at least one predictor and one target for the data, the data including a plurality of values of the predictor and a plurality of values of the target, the method comprises constructing a binary tree for the predictor that splits the values of the predictor into a plurality of portions, pruning the binary tree, and defining as bins of the predictor leaves of the tree that remain after pruning, each leaf of the tree representing a portion of the values of the predictor.

The invention discloses a product packaging surface defect detection and classification method based on machine vision. The method comprises the steps of 1, acquiring a high-definition color image of defect-free product packaging, making the high-definition color image into a standard image, conducting real-time shooting with a camera, conducting online acquisition of a high-definition color image of product packaging to be detected, and making the high-definition color image as an image to be detected; 2, conducting image matching on the image to be detected and the standard image based on SURF algorithm; 3, conducting difference image operation on the two images matched in step 2 to obtain a defective image; 4, conducting feature extraction on the defective image to obtain the geometrical features and color features of the defective image; 5, classifying product packaging surface defects by means of RBF neural network algorithm. Automatic defect detection and classification are conducted by means of a machine vision system, human factor interference can be avoided, labor cost is reduced greatly, and then huge hidden cost caused by training and management when artificial detection is adopted is avoided.

The invention relates to a mapping knowledge domain embedding method for fusing multiple background knowledge. The method includes the steps of firstly, selecting high-quality entity depiction information from entity labels of a knowledge base and selecting high-quality linguistic data related to an entity from Web linguistic data to establish the MCK (multiple background knowledge); secondly, learning the embedding expression of the knowledge base by embedding the MCK; thirdly, obtaining a semantic embedding vector of the corresponding entity from the MCK through a DBALSTM model, wherein DBALSTM represents depth D, bidirectional B, attention A and base LSTM; fourthly, completing the mapping knowledge domain embedding for the fusing multiple background knowledge by using a fusing embedding mechanism for the fine grit combination of three-element groups MCK and RDF. The accuracy of mapping knowledge domain embedding can be improved.

The invention discloses a clustering and reclassifying face recognition method, which comprises the steps of acquiring a training sample; carrying out equalization processing on the training sample; carrying out Gabor texture feature extraction on face images, and acquiring a feature vector corresponding to each face image after feature extraction; carrying out dimension reduction on acquired Gabor texture features of each face image to acquire feature vectors after dimension reduction; carrying out a clustering operation until distance convergence so as to complete clustering; classifying all of the clustered feature vectors to acquire a plurality of subclasses, calculating to determine each vector mean value, and calculating to acquire a within-class distance and an among-class distance; carrying out feature extraction and preprocessing on face images of a target to be recognized, acquiring a feature vector after projection transformation, and calculating the distance between the acquired feature vector and the feature vectors in each subclass sequentially so as to acquire the similarity; and determining identity information of the target to be recognized. The method disclosed by the invention can shorten the among-class distance so as to reduce an error in the acquisition process, and the accuracy of face recognition is improved.

A method for stabilizing a knowledge graph includes: generating a knowledge graph in which same entities in a semantic relation list between entities provided as an input are represented as a single node based on names and types of the entities; computing, on the knowledge graph, semantic similarities between all potential entity pairs of same entity types by comparing, for each potential entity pair, a type of relation associated with an entity in the entity pair and an opponent entity to the entity; and selecting, based on the semantic similarities, a representative entity from each of semantically similar entity pairs on the knowledge graph and integrating an opponent entity to the representative entity into the representative entity. The method further includes computing relation weighted values between the entities by using a graph analysis and statistic information, and adding the weighted values to the knowledge graph.

Implementations of the present principles include methods and systems for retransmitting un-recovered information within a cognitive radio, anti-jamming system. In accordance with aspects of the present principles, encoding schemes may be optimized for retransmission by utilizing a jamming rate and a number of un-recovered packets to minimize packet loss and thereby enhance throughput. In addition, rateless encoding features may be employed to re-encode un-recovered portions of an information sequence for efficient retransmission.

A flash ROM content updating method and system is disclosed, wherein the flash ROM has a plurality of memory blocks stored with primary core information. Check-sum calculation is performed on the primary core information to form and designate a check code to each memory block. Updated information is inputted and divided into a plurality of update information blocks corresponding in size to the memory blocks. The check-sum calculation is also performed on the updated information is performed to form and designate a check code to each update information block. When the check code of a memory block is not consistent with that of a corresponding update information block, the primary core information stored in the memory block is updated with the updated information in the corresponding update information block.

The invention discloses a Web instant messaging method which comprises the steps of: a, sending an instant message to a server by a first user; b, forwarding the instant message to a second gateway registered by a second user if the second user registers the Web, or storing the instant message into a database in an off-line message way if the second user does not register the Web; c, returning the off-line message to a client end of the second user through the second gateway when the second user is changed into the register state from the non-register state, and accessing the second gateway by the client end of the second user to obtain the instant message when the second user is in the register state; and d, if getting the instant message, displaying the instant message to the second user, if not, returning to the step c until the second user leaves the Web. By updating the state information of the user state, the Web instant messaging method solves the problem of delay of managing the on-line state of the user, and greatly lowers the possibility of information loss and delay in processes of sending and receiving the message.

The invention provides a credit scoring model construction method and a device, equipment and a storage medium. The construction method comprises the steps of dividing an original unbalanced credit data set into a training set and a verification set; wherein a plurality of data samples in the original unbalanced credit data set comprise credit information of a plurality of users, and the pluralityof data samples are in one-to-one correspondence with the plurality of users; dividing the data samples in the training set into a majority class of training samples and a minority class of trainingsamples; clustering the majority class training samples by using an unsupervised clustering algorithm to generate a plurality of sample clusters; obtaining a preset number of balance training subsetsaccording to the plurality of sample clusters and the minority class training samples; and constructing a credit scoring model according to the obtained balance training subset, the verification set and a preset decision tree base classifier. According to the method, the classification performance of the credit scoring model can be improved.

A content-adaptive video preview system (100) allows to go faster through a video than existing video skimming techniques. Thereby, a user can interactively adapt (S1) the speed of browsing and / or the abstraction level of presentation.According to one embodiment of the invention, this adaptation procedure (S1) is realized by the following steps: First, differences between precalculated spatial color histograms associated with chronologically subsequent pairs of video frames said video file is composed of are calculated (S1a). Then, these differences and / or a cumulative difference value representing the sum of these differences are compared (S1b) to a predefined redundancy threshold (S(t)). In case differences in the color histograms of particular video frames (302a-c) and / or said cumulative difference value exceed this redundancy threshold (S(t)), these video frames are selected (S1c) for the preview. Intermediate video frames (304a-d) are removed and / or inserted (S1d) between each pair of selected chronologically subsequent video frames depending on the selected abstraction level of presentation. Thereby, said redundancy threshold value (S(t)) can be adapted (S1b′) for changing the speed of browsing and / or the abstraction level of presentation.

The invention relates to the field of artificial intelligence, and discloses a voiceprint recognition method for original voice, which is used for reducing information loss and system complexity of anoriginal voice data recognition model of a speaker. The method comprises the steps of obtaining original voice data, and performing segmentation processing on the original voice data according to a preset time length to obtain segmented voice data; performing tail biting convolution processing and discrete Fourier transform processing on the segmented voice data through a preset convolution filter bank to obtain voiceprint feature data; pooling the voiceprint feature data through a preset deep neural network to obtain target voiceprint features; performing embedded vector conversion processing on the target voiceprint feature to obtain a corresponding voiceprint feature vector; and calculating the voiceprint feature vector through a preset loss function to obtain target voiceprint data, the loss function comprising a cosine similarity matrix loss function and a minimum mean square error matrix loss function. The invention also relates to a blockchain technology, and the voiceprint feature data is stored in the blockchain.

The invention discloses an interest recommendation method and system based on a user sequence click behavior, and the method comprises the steps: firstly obtaining the historical interaction project data of a user, and forming a historical interaction project sequence of the user; constructing an interest recommendation model; carrying out session division on the historical interaction item sequence of the user by utilizing an interest recommendation model; extracting interests in each session obtained after division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user; interacting interests among different sessions to obtain a dynamic interaction model among different sessions; and inputting the session interest sequence of the user into a dynamic interaction model between different sessions, and predicting and obtaining a to-be-recommended target project sequence.

This system consists of mobile terminals (MT), the safeguard control server (SCS) and the net-side database (ND). SCS controls the MT info operation. SCS connects directly to ND. MTs connect with SCS via wireless accessing net. This invention also discloses a MT info resource safeguarding method to spare the MT info. After MT is lost, MT info can be restored or the info in MT can be auto destroyed. These reduce effectively the loss of info losing or info leakage due to losing the handset.

Embodiments of the present application relate to a multilingual mixed search method, a system for multilingual mixed searching, and a computer program product for multilingual mixed searching. A multilingual mixed searching method is provided. The method includes receiving a query word entered by a user and identifying source language of the query word, translating the query word from the source language to query words of a plurality of target languages, the plurality of target languages being different than the source language, using the query word in a corresponding target language as a basis for searching an index of web page information corresponding to the corresponding target language and obtaining web page information in the corresponding target language, from the web page information obtained from the index corresponding to the corresponding target language, and translating the obtained web page information from the corresponding target language to the source language.

The invention discloses a discrete Bayesian network water chilling unit fault diagnosis method based on information entropy, and the method comprises the steps: obtaining the historical data of a fault through the historical data stored in an experiment or on site, and carrying out the steady-state screening and feature selection; carrying out discretization processing on historical data by utilizing an information entropy-based discretization algorithm, carrying out statistics on frequency to determine conditional probability, and constructing a network model; and verifying the performance ofthe model. According to the method, the main limitation of traditional Bayesian network water chilling unit fault diagnosis based on expert discretization is effectively overcome, and the possibilityof field application of a fault diagnosis system is greatly improved.

The present invention discloses a novel method and system for identification and analysis of fundamental stock characteristics, as well as investment funds distribution. Essential stock performance factors are identified and characterized, the identified factors are then divided by sectors. The weight on the stock selection decision of each factor is then identified based, in part, on how indicative the factor is of actual stock performance. Stocks are then purchased according to this selection method, and the unique fund division method of the present invention.

The invention discloses a short text similarity calculation system and a training method thereof. The system comprises the following modules: a text segmentation module, a text encoder and a text similarity calculation neural network module. According to the short text similarity calculation system, the same encoder is used for encoding two short texts of which the similarity needs to be calculated, then the attention mechanism is used for obtaining the attention of the first text to the second text, and the value of the similarity is obtained by normalizing the attention. According to the neural network, semantic coding of words in the text in the context and semantic coding of the whole text are effectively utilized, attention is used for representing similarity, information loss of text-level semantic coding is relieved, and the problem that word-level semantic coding is isolated from the context is avoided.

The invention discloses a text abstract generation method based on advanced semantics. The text abstract generation method comprises the following steps: (1) carrying out word segmentation on text corpora and converting the text corpora into semantic tag sequences in one-to-one correspondence with vocabularies; (2) on the text abstract model, using a bidirectional circulation network as an encoderto encode the vocabulary sequence and the semantic tag sequence, and abstract representation on vocabularies and abstract representation on semantics are obtained; (3) combining the abstract representation on the vocabulary with the abstract representation on the semantics; (4) sending the merged abstract representation into a decoder, respectively calculating vocabulary attention weight and semantic attention weight, and predicting probability distribution of each step of the sequence on a word list; and (5) combining the attention weight distribution and the word list probability distribution to obtain final output probability distribution, converting the final output probability distribution into readable vocabularies, and connecting the readable vocabularies in series to form sentences for outputting. According to the method, the accuracy of predicting the low-frequency words and carrying out the text abstract on the unlabeled data by the model can be improved.

An image processing apparatus that processes an input image to generate an output image to be recorded on a recording medium includes a first obtaining unit configured to obtain a size of the input image, a second obtaining unit configured to obtain a size of the recording medium, a setting unit configured to set, in two directions orthogonal to each other, an overrunning amount caused if the input image were recorded on the recording medium without a margin, a calculating unit configured to calculate a scaling factor in each of the two directions using the size of the recording medium, the overrunning amount, and the size of the input image, and a generating unit configured to perform a scaling operation on the input image in the two directions based on the respective scaling factors to generate an output image.

The present invention describes a road terrain detection system that comprises a method for classifying selected locations in the environment of a vehicle based on sensory input signals such as pixel values of a camera image. The method comprises a high level spatial feature generation for selected locations in the environment called base points. The spatial feature generation of the base points is based on a value-continuous confidence representation that captures visual and physical properties of the environment, generated by so called base classifiers operating on raw sensory data. Consequently, the road terrain detection incorporates both local properties of sensor data and their spatial relationship in a two-step feature extraction process.

The invention discloses a method for removing nuclear magnetic artifacts in an EEG signal based on automatic ICA. The method comprises the following steps that 1, basic denosing processing is conducted on the EEG signal comprising the nuclear magnetic artifacts, and an EEG signal X' (t) of which most of nuclear magnetic artifacts are removed is obtained; 2, independent element separation is conducted on the EEG signal X' (t), remaining elements of the nuclear magnetic artifacts in the EEG signal X' (t) are automatically identified, and other elements keep unchanged; 3, efficient low-frequency elements in remaining elements of the nuclear magnetic artifacts in the step 2 are extracted and reserved; 4, the efficient low-frequency elements in remaining elements of the nuclear magnetic artifacts obtained in the step 3 and other elements reserved in the step 3 are together reconstructed through ICA inverse transformation, and a denosed EEG signal is obtained. On the basis of the ICA, according to frequency distribution characteristics of nuclear magnetic noise and periodic characteristics of nuclear magnetic noise related to repetitive time parameters of nuclear magnetic scanning, the elements of the nuclear magnetic artifacts are automatically selected and the useful information is reserved.

A method reads out X-ray information stored in a storage phosphor layer where an X-ray picture is obtained and made up from a number of pixels. The method includes the steps of: irradiating the storage phosphor layer with a stimulation light beam which is moved along a line over the storage phosphor layer to stimulate the storage phosphor layer into emitting emission light; collecting emission light emitted from the storage phosphor layer, and converting the emission light collected into a corresponding detector signal S; deducing a number of detector signal values D from the detector signal S, where the number of deduced detector signal values D is greater then a predetermined number of pixels of the X-ray picture along the line; and deducing a picture signal value B for each pixel of the X-ray picture along the line from at least two of said detector signal values D.

The invention discloses a multi-source image target association method based on improved dictionary learning, belongs to the field of information data processing, and mainly solves the problems of information loss and redundant steps caused by first identification and then association of image target association in existing multi-source information data fusion. The method comprises the steps of firstly collecting image data sets of multiple information sources for the same scene or the same target to form an original data set; carrying out unified sparse representation on the multi-source image by utilizing an improved dictionary learning method, and increasing the representation discrimination capability of dictionary set characteristics by introducing label information into an objectivefunction; and constructing a neural network to learn the sparse representation and the label information of each image to obtain a distance measurement standard between the associated target and the non-associated target, and replacing a traditional distance measurement mode to complete the establishment of an association discrimination model. The method fully utilizes the feature information of the image, avoids the step redundancy of the existing method, and has the advantages of high model generation speed, less information loss, good practical effect and the like.

An apparatus and associated method for reading out X-ray information of an X-ray picture stored in a phosphor layer includes: an irradiation device for irradiating the phosphor layer with a stimulating light beam to be moved along a line over the phosphor layer, and in so doing stimulating the phosphor layer into emitting emission light; a detector for collecting the emission light emitted from the storage phosphor layer and for converting the emission light collected into a detector signal S; a processing unit for deducing picture signal values B for pixels of the picture along the line from the detector signal S; and two or more sensors located at different reference positions to collect reference times when the light beam is located at the different reference positions. The processing unit deduces a number of detector signal values D from the detector signal S to deduce the picture signal values B between the two reference times.