69 results about "Local invariant" patented technology

The invention discloses a monocular vision / inertia autonomous navigation method for an indoor environment, belonging to the field of vision navigation and inertia navigation. The method comprises the following steps: acquiring feature point information based on local invariant features of images, solving a basis matrix by using an epipolar geometry formed by a parallax generated by camera movements, solving an essential matrix by using calibrated camera internal parameters, acquiring camera position information according to the essential matrix, finally combining the vision navigation information with the inertia navigation information to obtain accurate and reliable navigation information, and carrying out 3D reconstruction on space feature points to obtain an environment information mapto complete the autonomous navigation of a carrier. According to the invention, the autonomous navigation of the carrier in a strange indoor environment is realized with independent of a cooperative target, and the method has the advantages of high reliability and low cost of implementation.

Systems and methods for utilizing relation- and query-specific information to specialize DBMS code at runtime based on identifying runtime locally invariant variables. Runtime invariant is often of the form of variables in code that hold values that are constant during a portion of code execution. Micro-specialization is applied to eliminate from the original program unnecessary code such as branching statements that reference local invariant(s) in branch-condition evaluation. The resulting specialized code reduces the code complexity as well as significantly improves the runtime efficiency during code execution.

The invention discloses a method for automatically tagging animation scenes for matching through comprehensively utilizing an overall color feature and local invariant features, which aims to improve the tagging accuracy and tagging speed of animation scenes through comprehensively utilizing overall color features and color-invariant-based local invariant features. The technical scheme is as follows: preprocessing a target image (namely, an image to be tagged), calculating an overall color similarity between the target image and images in an animation scene image library, and carrying out color feature filtering on the obtained result; after color feature filtering, extracting a matching image result and the colored scale invariant feature transform (CSIFT) feature of the target image, and calculating an overall color similarity and local color similarities between the matching image result and the CSIFT feature; fusing the overall color similarity and the local color similarities so as to obtain a final total similarity; and carrying out text processing and combination on the tagging information of the images in the matching result so as to obtain the final tagging information of the target image. By using the method provided by the invention, the matching accuracy and matching speed of an animation scene can be improved.

The invention provides a method for detecting an image spam email by local invariant features of pictures, which can extract the invariant region feature of junk information in the pictures by using a scale-invariant feature conversion algorithm and extract characters embedded into the pictures to classify the pictures so as to form a feature vector library of the pictures combining two features together. Experiments prove that the recall rate of the spam email can be improved and the program operation time and space can be saved. The method can extract the invariant region feature in the pictures to generate the feature vectors of the pictures, and a support vector machine classifier is used for training and testing. In the method, by utilizing the text messages embedded into the pictures, the text string in the pictures can be excavated by using a graphic character recognition technology and the string can be taken as the feature of the pictures, and the Bayesian classifier is used for training and testing. The feature vector of each picture is composed of the local invariant feature of the picture and the text string; and two types of classifiers are used for classifying by a stacking method to achieve the purpose of detecting the image spam email.

The invention discloses a high-efficiency method and a high-efficiency system for sensitive image detection. The method comprises that sensitive image samples and normal image samples are collected to establish a training set and to extract interest points, the interest points are filtered in combination with a skin color model, the interest points unrelated to skin colors are taken out and the interest points related to the skin colors are kept, local invariant characteristics at the interest points are extracted and are clustered, a data-driven tree pyramid model is established, and the multi-resolution histogram characteristics of each image are extracted on the basis; the similarity of any two images is calculated by using pyramid matching algorithm and a kernel function matrix is formed; and the obtained kernel function matrix is used to train a support vector machine classifier to obtain the parameters of the classifier and a new image sample is detected to determine whether the new image sample is a sensitive image. The invention can conduct high-efficiency detection and filtration to the sensitive images on the internet to enable the vast juvenile to enjoy the convenience brought by the internet and to protect the vast juvenile against the harmfulness of bad information.

The invention discloses a semi-supervised speech feature variable factor decomposition method. Speech features are divided into four types: emotion-related features, gender-related features, age-related features and noise, language and other factor-related features. Firstly, a speech is pretreated to obtain a spectrogram, speech spectrum blocks of different sizes are inputted to an unsupervised feature learning network SAE, convolution kernels of different sizes are obtained through pre-training, convolution kernels of different sizes are then respectively used for carrying out convolution on the whole spectrogram, a plurality of feature mapping pictures are obtained, maximal pooling is then carried out on the feature mapping pictures, and the features are finally stacked together to form a local invariant feature y. Y serves as input of semi-supervised convolution neural network, y is decomposed into four types of features through minimizing four different loss function items. The problem that the recognition accuracy rate is not high as emotion, gender, age and speech features are mixed is solved, and the method can be used for different recognition demands based on speech signals and can also be used for decomposing more factors.

The invention relates to a method for tracking a target based on graph theory cluster and color invariant space. The method provided by the invention comprises the following steps of: (1) carrying out color invariant space transformation on an image of a video flow; extracting feature points by utilizing color invariant features CSIFT (Colored Scale Invariant Feature Transform); (2) carrying out graph theory movement cluster on the features, wherein the feature points with the same movement tendency belong to targets with the same movement state; and tracking targets in a video frame by the feature points. In the invention, local invariant features are utilized and the blindness of overall features is abandoned; the extraction of the color invariant features CSIFT and increases the colorinvariance property is realized while the geometrical invariability advantages of the SIFT features are kept so that a gray level feature space is advanced to the color space; to a movable object under a static background, the graph theory through is matched and a feature movement classification is used as the pre-treatment of matching so that the matching operation is more accurate and faster.

The invention discloses a method for matching a generalized Hough transform image based on local invariant geometrical characteristics to achieve the purpose of matching a target image with any rotating angle. The method is characterized by comprising the step of preprocessing a template image and the step of matching the target image through the processed result of the template image. The step of processing the template image comprises the step of extracting all edge points of the image and establishing an edge point matching characteristic relation. When the matching process is carried out on the target image, the edge points of the target image are extracted, the extracted image is matched through the established edge point matching characteristic relation, accordingly, the matching position points and the corresponding rotation angles are acquired, and the target image is matched. A traditional Hough transform is improved, an improved reference table is established, the target image rotated by any angles can be matched, and a very high matching speed and matching precision are achieved.

The invention discloses a traffic target tracking method based on optical flow and local invariant features. The method comprises the steps that firstly, an initial template is constructed for an input video image through Gaussian background modeling, and a foreground target is extracted; next, a target characteristic point is detected through the SURF transformation algorithm; then, the characteristic point is detected and the target is tracked by constructing an image multi-resolution wavelet pyramid and improving an LK sparse optical flow method, an adaptive template real-time updating strategy is made, whether the template is the last frame or not is judged, and if yes, tracking is ended; if the template is not the last frame, template updating judgment is conducted; if template updating is not needed, tracking is continued; if template updating is needed, tracking is continued after the template and a tracking window are updated according to an updating method. Through the method, matching is accurate and rapid, redundant data is reduced, adaptability is high, high robustness is achieved in respect of target vehicle deformation, high speed, noise, uneven illumination, partial covering and other complicated environments, and the vehicle recognition ability is improved; the method has obvious advantages compared with a traditional method and has good application prospects in intelligent transportation target tracking systems.

The invention is applicable to the technical field of image processing and provides a semantic mapping method of the local invariant feature of an image. The semantic mapping method comprises the following steps of step A: extracting and describing the local invariant feature of the colorful image; step B: after extracting the local invariant feature, generating a visual dictionary for the local invariant feature extracted from the colorful image on the basis of an algorithm for supervising fuzzy spectral clustering, wherein the visual dictionary comprises the attached relation of visual features and visual words; step C: carrying out semantic mapping and image description on the attached image with the local invariant feature extracted in the step A according to the visual dictionary generated in the step B. The semantic mapping method provided by the invention has the advantages that the problem of semantic gaps can be eliminated, the accuracy of image classification, image search and target recognition is improved and the development of the theory and the method of machine vision can be promoted.

The invention discloses an identify recognition method and an identify recognition system. First, a high-resolution palm vein image is acquired through an image acquisition mode combining CCD and FPGA. Then, the original palm vein image is preprocessed, local invariant features of the image are extracted based on training data (acquired during registration) and test data (verified online), the similarity between a test data feature point vector and a training data feature point vector is measured through use of Euclidean distance for feature matching, a decision is made according to the feature matching rate after matching, a decision result is output directly for a palm vein image with high feature matching rate, and for a palm vein image with low feature matching rate, image 3D deflection angle estimation and 3D rotation are carried out, feature selecting and matching is carried out again on the rotated image, and a decision result is then output directly.

The invention discloses a video image stabilization method based on color constant and geometry invariant features. The method comprises the steps of based on the color pattern transformation, building a multi-scale space based on the color constant mode, extracting local invariant feature points based on the color constant and geometry multi-scale, conducting feature point matching and shaking movement estimation on a video image sequence, then calculating the affine transformation matrix of adjacent frame images, conducting transformation on a current frame relative to deformation of a reference frame image through a cumulative affine transformation matrix, achieving motion compensation, finally achieving video image stabilization, removing abnormal motions, such as translation, rotating and scaling between the images, generated by shaking of a camera, of the video images automatically in real time, removing video shaking caused by vibration of the camera, and therefore providing the stable video images.

The invention discloses a contour-based local invariant region detection method. The method mainly utilizes contour corners, angle bisectors of the contour corners and feature points on a contour invariant relative to the angle bisectors to construct an invariant region. Because the angle bisectors have strong anti-noise capability and the affection of rotation, scale and other factors on the angle bisectors is little, the region obtained by the method has high stability and repeatability, and repetition rate experiments of rotation, scale, affine, illumination, noise, blur and the like provethat the method has high processing speed, strong robustness and wide applicability.

The invention discloses a method for detecting a spam email by utilizing an improved gauss hybrid model classifier, comprising the following steps of: extracting invariant region features of spam information in a picture by utilizing an accelerative extract algorithm of a robust feature, executing the fitting of a gauss hybrid model on the invariant region features, and executing the evaluation of weight, mean and covariance matrixes by using an expectation maximization method, wherein the method specifically comprises the following steps of: labeling pictures of a data set to be detected, and dividing the pictures into spam pictures and regular pictures; extracting vectors of local invariant features of all data sets by utilizing the accelerative extract algorithm of the robust feature; executing density function fitting on the local invariant features by utilizing the gauss hybrid model to obtain mean and covariance matrixes of the all pictures; improving a mean clustering algorithm to make the mean clustering algorithm be suitable for clustering special feature vectors obtained in the previous step, taking cross entropy as an measurement index of the similarity of gauss hybrid distributions, and realizing the mean clustering algorithm based on the gauss hybrid model; and establishing a classifier by utilizing the mean clustering algorithm based on the gauss hybrid model.

The invention discloses an image tracking method for a strapdown optical seeker. According to the method, influence caused by severe blurring and degradation of imaging in a missile flying process is inhibited by adopting motion-blurred image restoration and a local characteristic based target tracking technology. The method comprises steps as follows: a target characteristic template is extracted firstly, a multi-scale Harris characteristic point and an SIFT descriptor of a local invariant characteristic area of the characteristic point are acquired, and a to-be-tracked characteristic point is selected; secondly, a follow-up frame of image is subjected to motion-blurred image restoration; thirdly, a gate position is predicated through transformation of coordinates on the basis of a missile attitude angle, so that image processing data is reduced, and real-time performance is improved; finally, vector matching of the SIFT descriptor is performed in a gate, and the optimal matching characteristic point is determined, and a target template is updated. According to the method, a target can be tracked stably when the target imaging is severely blurred and degraded due to missile jittering.

The invention discloses a vehicle target detection method and system based on YOLOv2, which comprises the steps of obtaining sample traffic video data; dividing the sample traffic video data into frame images to serve as sample images; performing noise reduction, shadow elimination, local histogram equalization and local invariant analysis on the sample images to obtain training images; inputtingthe training images into a preset YOLOv2 neural network model for training to obtain a vehicle detection model; obtaining real-time vehicle video data; performing noise reduction, shadow elimination,ghost elimination, local smoothing and local invariant analysis on the real-time vehicle video data to obtain secondarily processed real-time vehicle video data; segmenting the secondarily processed real-time vehicle video data into frame images, and inputting the frame images into the vehicle detection model to obtain a result map. The vehicle detection accuracy and the detection speed can be improved according to the vehicle target detection method and system based on YOLOv2 provided by the invention.

The invention discloses an adaptive mean shift algorithm based on local invariant feature detection. According to the adaptive mean shift algorithm based on local invariant feature detection, local invariant feature detection and the adaptive mean shift algorithm are combined, the detection and matching of local invariant feature points of an object are introduced during searching, and the region of search is recalculated through obtained matched feature points, so that the region of search can be excellently constrained around a target range, and finally, the accuracy of a tracking process is ensured. The adaptive mean shift algorithm based on local invariant feature detection has the advantage that the accuracy and stability of searching are greatly improved relative to those of adaptive mean shift algorithms.

The invention relates to a method for personnel trace extraction by utilizing image collecting and processing technology. The method mainly includes the steps of image collection, image registration, interested area creating, differential signal thresholding and ecological processing, wherein positioning detection and description of feature points are realized by utilizing a graphic local invariant feature extracting method during image registration; an interested area is created by utilizing a gradient field and a direction field of an integral image to determine trace texture and background image and then utilizing gray level; differential image thresholding refers to solving a maximum threshold value of the interested area through an OSTU algorithm and binarizing according to the maximum threshold value; ecological processing is performed finally. By the personnel trace extraction method, operation quantity can be reduced, and extraction accuracy can be improved; the personnel trace extraction method is suitable for various personnel traces, thereby being high in applicability.

The invention discloses a gesture recognition method based on address event flow characteristics. The gesture recognition method is mainly used for solving the gesture recognition problem under a complex background. The implementation scheme comprises the following steps of (1) collecting the address event flow data; (2) de-noising each address event flow sequence; (3) confirming a peak address event flow sequence; (4) detecting a characteristic event of the peak address event flow sequence; (5) extracting local invariant features of the feature event; (6) screening local invariant features ofthe effective gesture; (7) training a support vector machine SVM classifier; (8) classifying;. According to the method, the asynchronous characteristic of the address event is reserved, non-effectivegesture characteristic calculation is reduced, and only the characteristic event is subjected to characteristic extraction. The method has the advantages of high accuracy and strong applicability.

The present invention provides a video-based palm print and palm vein joint registration and recognition method. The method includes a registration method and a recognition method. According to the registration method, images are extracted from a registration palm vein video and a registration palm print video; registration palm vein template features, registration palm vein LBP features and registration palm print local invariant features are obtained; and the registration palm vein template features, registration palm vein LBP features and registration palm print local invariant features are stored in a registration database. According to the recognition method, images are extracted from a recognition palm vein video and a recognition palm print video; recognition palm vein template features, recognition palm vein LBP features and recognition palm print local invariant features are obtained; and the recognition palm vein template features, the recognition palm vein LBP features and the recognition palm print local invariant features are matched, so that whether a user has been registered is recognized. With the joint registration and recognition method of the invention adopted, the palm recognition of a motion video can be realized, the user friendliness of the recognition can be effectively enhanced; a new strategy according to which a palm rotation video and a palm cross sweep video are registered in a fused manner is provided, and therefore, the richness and completeness of registration features can be improved, the robustness of the method to different recognition attitudes is improved; and a cascade fusion strategy is provided, and therefore, the recognition speed of registered users can be greatly improved.

The invention discloses a land mobile distance-based image spam similarity-detection method. Invariant area features of a spam are extracted from a picture by utilizing a scale invariant feature transform algorithm, and similarity between the picture to be detected and the picture in a spam feature library is calculated by using a land mobile distance, thereby detecting an image spam. In the land mobile distance-based image spam similarity-detection method provided by the invention, the local invariant features of the picture are utilized. In the prior art of detecting the spam by utilizing the similarity, a Euclidean distance is mainly utilized, cannot process features with variant structure sizes, and is required to cluster and normalize the features first, so the detection speed is influence. The local invariant features with the variant structure sizes are directly processed by utilizing the land mobile distance, so the method greatly increases the detection speed of the image spam and simultaneously ensures high accuracy and low false rate.

The invention relates to a method for detecting image-based spam by utilizing the image local invariant feature, which comprises the steps of extracting invariant region feature of spam information in an image by utilizing the accelerated extraction algorithm with the robust feature, further generating a feature vector of the image, estimating parameters of a Gaussian mixture model by using the maximum likelihood algorithm and training a classifier of the Gaussian mixture model. Experiments show that the method can improve the recall rate of the spam and save the program computation time and the space. The classifier based on the Gaussian mixture model is obtained. The realizing method for detecting the image-based spam comprises three modules of the extraction of the image feature, the estimation of the parameters of the Gaussian mixture model and the detection of the image-based spam.

The invention discloses a view landmark retrieval method based on end-to-end deep learning, and the method comprises the following steps: S1, collecting a key landmark image, carrying out the preprocessing operation, and enabling the key landmark image to serve as training data; S2, embedding a local aggregation descriptor feature vector method into the CNN to form an end-to-end CNN model; S3, inputting the collected training data into an end-to-end CNN model, extracting image local invariant features, training the CNN model through an error function, and learning an optimal aggregation cluster center point; S4, performing key frame picture extraction operation on the to-be-identified video stream, and performing down-sampling operation after the to-be-identified video stream and the to-be-identified picture stream are subjected to the down-sampling operation to generate a to-be-identified landmark data set Q; S5, inputting Q into the trained CNN model, performing local invariant feature vector extraction, and outputting a calculation result of each landmark category through a full connection layer and a data output layer; And S6, according to a key landmark category threshold value set by training, judging whether each piece of data in Q has a key landmark category or not, and if yes, outputting a picture source name and landmark prompt.

The invention belongs to the data recognition and data representation technical field and discloses a local invariant gray feature-based image registration method and an image processing system. According to the local invariant gray feature-based image registration method, feature extraction descriptors are constructed; feature points between registration images are searched, the nearest neighborprinciple is used to find matched key points; and an affine transformation matrix H between the registration images is calculated, and six parameters of the affine transformation matrix H are obtainedthrough singular value decomposition. The descriptors are constructed; sampling points are divided into an odd part and an even part, and therefore, dimensionality during the construction of the descriptors is significantly lowered, operating time is reduced, the accuracy and accuracy of registration are improved; when being constructed, descriptor vectors are sequenced according to gray values,and therefore, rotation invariance can be realized. The method of the invention has high detection precision, good noise robustness and low computational complexity, which mainly benefits from the great reduction of the dimensionality of the original descriptors and insensitiveness to illumination transformation.

The invention discloses a method and device for obtaining the affine local invariant features of an image. In order to solve the problem of extraction of local invariant features of machine vision, the method comprises the following steps: obtaining a target image including a plurality of pixel points; establishing a local feature conversion model corresponding to each pixel point; according to each local feature conversion model, further obtaining a local direction tensor corresponding to each pixel point; according to each local direction tensor, determining the maximum value of the smaller feature value of each pixel point in a preset area corresponding to each pixel point to further determine the pixel point corresponding to the maximum value as an initial interesting point; utilizing an affine recursion algorithm to converge each initial interesting point into an affine interesting point and an affine feature area to obtain an image coordinate and feature scale of the affine interesting point and the affine feature area corresponding to the affine interesting point, so that the affine local invariant features of the image are obtained.

The present invention discloses an article antitheft detection method based on visual tag identification, which performs local feature matching for a frame image extracted from a video and a database image to achieve antitheft detection, thereby improving speed, reliability and accuracy. The article antitheft detection method overcomes the problems that a conventional video article antitheft detection method has low reliability in background modeling and motion segmentation of a video sequence and has a low accuracy in article identification under a complex environment condition. The article antitheft detection method comprises the implementation steps of: (1) extracting local features of a detected article, and establishing a visual tag database; (2) extracting a frame image from a video stream at a fixed time interval; (3) extracting the same local feature, matching the local feature with the local features in the visual tag database, and removing false matching point pairs; and (4) judging whether the number of matching points exceeds a threshold. The article antitheft detection method of the present invention does not need sequence information and only needs a single frame image to perform article antitheft detection, thereby improving detection speed; in addition, local invariant feature matching achieves detection and identification on the condition that the article is partially shielded or illumination changes, thus detection accuracy is ensured.

The invention relates to a high-resolution remote sensing image registration method based on a local invariant feature point. The high-resolution remote sensing image registration method comprises the following steps: S1: extracting Harris feature points in a benchmark remote sensing image and a remote sensing image to be registered at different time phases in the same area to independently obtain feature point sets P1 and P2; S2: utilizing a SIFT (Scale Invariant Feature Transform) descriptor to independently carry out feature vector description on the feature point sets P1 and P2; S3: searching bidirectional matching point pairs; S4: randomly selecting three groups of matching point pairs PM3 from S3, and obtaining the root-mean-square error RM of the three groups of matching point pairs PM3; S5: judging the threshold value, and returning to S4 or entering S6; S6: calculating an affine transformation relationship matrix Matrix; and S7: utilizing transformation in S6 to obtain a registration image image_R. The high-resolution remote sensing image registration method solves the problem of big registration error of the high-resolution remote sensing image, can realize the high precision and the automation of registration and has a wide application value in the field of the change detection of the remote sensing image.

The invention discloses a robot and an inventory management method thereof. The invention management method comprises the steps that a front image of a to-be-recognized object is collected, and firstlocal invariant features of the to-be-recognized object are extracted according to the front image; an environment image is acquired, and environment local invariant features of the environment imageare extracted; the environment local invariant features are matched with the first local invariant features, matching feature points are mapped to a reference center to obtain an effective reference center, and a scale ratio and a reference clustering radius corresponding to the effective reference center are acquired; second local invariant features of the front image of the to-be-recognized object are extracted again according to the scale ratio, then the environment image is matched with newly extracted the second local invariant features, and feature points obtained after matching are mapped to the reference center; and finally clustering is performed on the reference center of the object according to the reference clustering radius. In this way, the precision and speed of detection and recognition can be improved in terms of article recognition.

The invention discloses a hidden variable decoupling-based face image local feature migration network and method. The method comprises the steps of firstly proposing a hidden variable decoupling-basedlocal makeup migration and removal generative adversarial neural network framework; on the basis of a network structure, and innovatively providing a local makeup migration loss function which can beused for training the network, wherein the loss function comprises a local makeup loss function, a local invariant loss function and a local and global makeup invariant loss function; regarding the face picture which is not made up as a special case of the face picture which is made up, processing makeup migration and makeup removal as a unified problem, and obtaining multiple makeup removal effects.