62 results about "Bhattacharyya distance" patented technology

The invention discloses an automatic fast segmenting method of a tumor pathological image. The method comprises the following steps: firstly filtering a tumor original pathological image through the adoption of a Gaussian pyramid algorithm to respectively obtain pathological images with equal resolution, double resolution, fourfold resolution, eightfold resolution and 16-fold resolution; determining an initial region of interest containing the tumor on the equal resolution image through a RGB color model and morphological close operation; iteratively optimizing the initial regions of interest from the equal resolution to the fourfold resolution through the adoption of bhattacharyya distance; judging that the contribution of the RGB color model to the tumor region of interest has been reduced to zero when the bhattacharyya distance achieves a set threshold value; performing the self-adaptive high resolution selection of the deep precise segmentation through the adoption of a convergence exponent filtering algorithm, thereby further segmenting under the most suitable high resolution; and finally segmenting out a normal tissue and a tumor tissue in the tumor region of interest through the adoption of a bag of words model based on random projection. The method disclosed by the invention has the features of being accurate, fast and automatic.

The invention discloses a target tracking method based on multi-feature self-adaption fusion and on-line study. Target features are extracted and used as template features; three types of features are extracted from each of novel candidate target regions; a self-adaption fusion process is carried out according to distinctiveness and correlation of all the features; Bhattacharyya distances between the features obtained after fusion and the template features are calculated; the Bhattacharyya distances are subjected to uniformization and then used as weights of the novel candidate target regions; the novel candidate target region with the maximum weight, and a target region are subjected to an overlap judgment, if the overlap rate is less than an overlap rate threshold value, a multiple-time region of the novel candidate target region with the maximum weight is input to a detector, and when a recognizer outputs 'yes', the fact that target tracking succeeds is shown, and the recognizer, the template features and the target region are updated; if the recognizer outputs 'no', the fact that a novel target is found is shown; and if the overlap rate is greater than or equal to the overlap rate threshold value, the recognizer, the template features and the target region are updated. Through the target tracking method based on the multi-feature self-adaption fusion and the on-line study, the adaption ability of the target tracking under different scenarios and certain deformation conditions is enhanced, and the problem that tracking drifting is prone to occurrence after occlusion is avoided.

The invention discloses a video semantic scene segmentation method based on a convolutional neural network, which is mainly divided into two parts, wherein one part is that a convolutional neural network is built on the basis of shot segmentation, and then semantic feature vectors of video key frames are obtained by using the built convolutional neural network; and the other part is that the Bhattacharyya distance between the semantic feature vectors of two shot key frames is calculated by using the time continuity of the front and back key frames according to the semantic feature vectors, andthe semantic similarity of the shot key frames is obtained through measuring the Bhattacharyya distance. Probability estimation values of different semantics are outputted by using the convolutionalneural network to act as a semantic feature vector of the frame. Considering a time sequence problem of scene partition in the continuous time, the shot similarity is compared by combining semantic features of the two shot key frames and the time sequence feature distance between the shots, and thus a final scene segmentation result is obtained. The method disclosed by the invention has certain universality and has a good scene segmentation effect under the condition that training sets are sufficient.

A speech recognition device and a speech recognition method thereof are disclosed. In the speech recognition method, a key phrase containing at least one key word is received. The speech recognition method comprises steps: receiving a sound source signal of a key word and generating a plurality of audio signals; transforming the audio signals into a plurality of frequency signals; receiving the frequency signals to obtain a space-frequency spectrum and an angular estimation value thereof; receiving the space-frequency spectrum to define and output at least one spatial eigenparameter and, and using the angular estimation value and the frequency signals to perform spotting and evaluation and outputting a Bhattacharyya distance; and receiving the spatial eigenparameter and the Bhattacharyya distance and using corresponding thresholds to determine correctness of the key phrase. Thereby this invention robustly achieves high speech recognition rate under very low SNR conditions.

The invention relates to a laser radar echo signal de-noising method based on variational mode decomposition. The method comprises the steps of according to a laser radar equation, simulating a laserradar echo signal; selecting an optimal decomposition layer number K in variational mode composition by means of detrended fluctuation analysis; performing variational mode decomposition on the laserradar echo signal; then calculating similarity between the laser radar echo signal and each mode probability density function after decomposition through Bhattacharyya distance, differentiating a correlation mode and an uncorrelated mode; then processing the uncorrelated mode by means of a moving average mode, and extracting a related signal; and finally reconstructing the correlation mode and theprocessed uncorrelated mode for realizing de-noising of the laser radar echo signal. The laser radar echo signal de-noising method can effectively filter the noise in the laser radar echo signal andimproves effective detecting distance and precision of the laser radar.

The invention provides a pedestrian identification method based on image with FHOG-LBPH feature. By statistical average of fusion HOG features (FHOG) , the optimal features are selected according to the combination with single optimal feature and the separable criterion of bhattacharyya distance, and improved FHOG-LBPH feature are obtained through fusion LBPH characteristics, fundamentally reducing the feature dimension; A classifier is obtained by using a support vector machine (SVM) to train the sample characteristics in order to obtain the classification of the test sample. The experimental results show that the method makes the pedestrian detection accuracy and real-time performance a certain improvement. Effectiveness of the method is validated by the image automatically shot, and the method has certain application value in the real pedestrian recognition.

The invention relates to a method for extracting a high-resolution SAR image building zone through multi-feature weighted fusion. The method is mainly characterized in that the strategy that feature weights are determined through a distance formula is novel. The method includes the following steps that 1, intensity images are subjected to filtering and other preprocessing; 2, features are extracted based on a gray level co-occurrence matrix texture analysis method and a variation function texture analysis method; 3, the feature weights are determined based on the Bhattacharyya distance; 4, the features obtained in the step 2 are subjected to weighted fusion; 5, the fused feature images are subjected to unsupervised classification through a k-means clustering algorithm; 6, a classification result is post-processed, small zones are removed, gaps are filled, an external contour is extracted, and the building zone is obtained. The texture features facilitating building zone extraction are fully synthesized, an established fusion feature map contains texture information richer than that of a single-feature map, the classification accuracy of the building zone and a non-building zone is improved to a certain degree accordingly, the extraction accuracy of the building zone is improved, and finally the boundary contour of the extracted building zone better fits a real building zone contour.

The embodiment of the invention provides an image quality evaluation method and system. The method comprises the steps of: obtaining gray contrast of an image to be measured according to a Bhattacharyya distance between a grey level histogram of an area of interest and a grey level histogram of a peripheral area; obtaining first texture features according to average texture features of the area ofinterest after filtering and polarity texture features of the area of interest after filtering; obtaining second texture features according to average texture features of the peripheral area after filtering and polarity texture features of the peripheral area after filtering; obtaining texture contrast of the image to be measured according to a Bhattacharyya distance between the first texture features and the second texture features; and obtaining the probability of detecting a target in the image to be detected according to the gray contrast and the texture contrast. Starting from the gray features and the texture features of the image, the image quality evaluation method and system comprehensively consider the detection conditions of tiny target features and employ gray and texture features to enhance the robustness and the applicability of the image evaluation method.

The invention provides a UI test method for a control. The method comprises: intercepting a target picture and a source picture, and calculating a color histogram of the target picture and the source picture to obtain a Bhattacharyya distance; when the Bhattacharyya distance is greater than a predetermined threshold, obtaining positions and first pixel values of a plurality of mutation pixel points on the target picture and second pixel values of pixel points, in the positions, on the source picture, wherein the mutation pixel points are pixel points on a UI profile; and outputting a test result according to a variance or a standard deviation of difference values of the first pixel values and the second pixel values. The invention furthermore provides a UI test system for the control. According to the UI test method and system for the control, a UI of the control can be automatically verified to test the display effect of the UI.

The invention discloses a method for building a multi-classification support vector machine classifier based on the Bhattacharyya distance and a directed acyclic graph, belonging to the field of mode identification. The method aims to solve the problems that the training speed is lowered when the number of training samples or the number of classes is increased and the calculation amount is large because the traditional multi-classification strategy can not utilize prior information with different separabilities among classes because of the fixed structure. The method disclosed by the invention comprises the following steps: 1. respectively calculating the Bhattacharyya distance between every two classes in a training sample for a multi-classification object; 2. building an initial operation form according to the Bhattacharyya distance between the every two classes in the training sample, which is obtained in the step 1; 3. according to the initial operation form obtained in the step 2, building a multiple classifier based on the directed acyclic graph structure; and 4. adopting a support vector machine as a binary classifier, and carrying out multi-classification based on the directed acyclic graph structure.

The invention provides an automatic retrieval method of a key frame in an IVUS (Intravascular Ultrasound) video. According to the method, firstly, each frame of image is subjected to SRAD (Speckle Reducing Anisotropic Diffusion) filtering; then, on the basis of performing grayscale stretch on the filtered images, a grey level histogram of each frame of image is calculated to obtain a 256-dimension normalized grayscale feature vector; then, the Bhattacharyya distance between the adjacent frame of scale feature vectors is calculated to be used as the measure on the similarity of adjacent frames of images; and finally, the Bhattacharyya distance between the adjacent frame grayscale feature vectors is compared with a set threshold value, and the extraction of the key frame is completed. The automatic retrieval method has the advantages that the fast extraction of the key frame in the IVUS video is realized through analyzing the grayscale features of the IVUS images; a tester can be helped to fast locate the variation or foreign object implanted position; and reliable information is provided for the processing and the evaluation of vascular mutation.

The invention belongs to the technical field of radar technology and discloses a multi-station radar signal fusion detection method based on Bhattacharyya distance quantization, and the communication bandwidth between radars and a signal fusion center in multi-station radar signal fusion detection can be reduced. The method comprises a step of setting a multi-station radar system which comprises a signal fusion detection center and N radar stations, a step of determining the quantization bit b of each radar station to a received echo signal and obtaining a quantization interval number M=2b according to the quantization bit b, a step of carrying out quantization of the received echo signal by each radar station, determining the quantization interval to which a quantized signal belongs, and transmitting a label number corresponding to the quantization interval to the signal fusion detection center, and a step of setting an expected false alarm probability, wherein the signal fusion detection center detects a target according to the expected false alarm probability and the label number corresponding to each quantization interval sent by N radar stations.

The invention discloses a target tracking method for judging occlusion and repositioning a target, and belongs to the field of target detection and tracking. The method comprises the following steps:for any target tracking process, roughly judging whether occlusion occurs or not by utilizing the change of a confidence coefficient function; calculating the Bhattacharyya distance between the targetLocal Binary Pattern feature and the Local Binary Pattern feature of the pre-selected box for the area which is preliminarily judged to be shielded, and further confirming whether the target Local Binary Pattern feature is shielded or not; for the condition of local occlusion, adopting a multi-peak repositioning method to perform positioning after target occlusion; and for severe shielding or long-time shielding, adopting an SVM (Support Vector Machine) detection repositioning method to carry out accurate positioning. According to the method, the target shielding condition can be accurately judged, and the problem of target repositioning after shielding is effectively solved.

The invention relates to a human body motion data similarity automatic evaluation method. First of all, by use of human body skeleton structure based motion data acquired by a motion capture device, distances between attitude are defined based on a data format, then statistical information of motion is established by aggregating various time-space attitude, finally, a distance between two motions is calculated based on a Bhattacharyya distance, segmented motion comparison is performed, and motion similarity is evaluated through the distance. The method concretely comprises the following steps: 1, performing human body motion acquisition by use of a Vicon motion capture suite; 2, decomposing motion into sub-motion of a time sequence; 3, performing similarity comparison on the sub-motion; and 4, obtaining the similarity between motion A and motion B by calculating Cos t<*>(h0,h0). The method has the following advantages: by taking attitude data of motion capture as input, the statistical information of the motion is established by aggregating the various time-space attitude such that the motion evaluation accuracy is improved and the real-time performance is improved.