294 results about "Hash coding" patented technology

E-mail addresses are converted by a hashing function into hash codes. An e-mail mass mailer that can have millions of addresses hash codes all addresses to which a message might be sent to form a first list. An entity that wishes to exclude delivery of a message to various e-mail addresses, hash codes such addresses to form a second list. Hash codes in the first list that are also in the second list are purged from the first list to yield a third list of hash codes, each of which represents an e-mail address to which a message may be sent.

The invention discloses a cross-modal hash retrieval method and system for fusing supervision information. The method comprises the following steps: constructing an image network, a text network and afusion network; The training sample pairs of image and text features are obtained and input into image network and text network respectively. Taking the output characteristics of the image network and the text network as the inputs of the fusion network, and defining the output of the fusion network; Constructing an objective function for learning a unified hash code according to the output of the converged network and the similarity between pairs; Solving the objective function to obtain a unified hash code; The unified hash code is used as the supervisory information and the semantic information is combined to train the hash network of a specific mode. The invention simultaneously learns the feature representation and the hash coding based on the end-to-end depth learning framework, canmore effectively capture the correlation between different modal data, and is conducive to the improvement of the cross-modal retrieval accuracy.

The invention discloses a cross-model Hash retrieval method on the basis of cooperative training and matrix decomposition. By the aid of the cross-model Hash retrieval method, the similarity between models and the internal similarity of the models can be effectively constrained for unlabeled cross-model data. The cross-model Hash retrieval method includes implementation steps of acquiring original data and carrying out normalization processing on the original data; carrying out cooperative training to obtain constraints between the models; acquiring internal constraints of the models by the aid of neighbor relations; decomposing training data matrixes and adding the constraints between the models and the internal constraints of the models into the training data matrixes to obtain objective functions; carrying out alternate iteration to obtain expressions of basis matrixes, coefficient matrixes and projection matrixes; carrying out quantization to obtain Hash codes of training data sets and test data sets; computing the Hamming distances between every two Hash codes of the data sets; sorting the Hamming distances to obtain retrieval results. The cross-model Hash retrieval method has the advantages that constraints on the similarity between the models of the cross-model data can be obtained by the aid of cooperative training processes, accordingly, the image and text mutual retrieval performance can be improved, and the cross-model Hash retrieval method can be used for picture and text mutual search service of mobile equipment, internets of things and electronic commerce.

The invention provides a large-scale image library retrieval method based on the local similarity hash algorithm. The large-scale image library retrieval method includes the steps that a part of images are selected from an image library to be retrieved to serve as a training image set, and SIFT features of training images are extracted; a K means algorithm is used for conducting clustering on the SIFT features of the training image set to obtain a codebook; the inverse frequency of each code word in the codebook is calculated on the training image set; local sensitive hash coding is conducted on each code word; SIFT features of a queried image and images in the image library to be retrieved are extracted respectively; for each image, the word frequency of each code word in the corresponding image is calculated, and then the weight of each code word is obtained; local similarity hash codes of the images are calculated by using the similarity hash algorithm; the Hamming distances between a hash code of the queried image and the hash codes of the images to be retrieved are calculated; the Hamming distances are used for retrieving the images similar to the queried image rapidly. The large-scale image library retrieval method has good universality, reduces data storage space and also improves the query retrieval efficiency.

The embodiment of the invention discloses an image retrieval method, device and apparatus and a computer readable storage medium. The method comprises the steps of enabling image pairs in an image database to serve as input, the distance between Hash coding pairs obtained through mapping of the image pairs, enabling the label category and the feature similarity of the image pairs to serve as lossvalues, and adopting a machine learning optimization algorithm to optimize the loss values so as to obtain a deep Hash mapping model through training; mapping the to-be-retrieved image into a to-be-retrieved Hash code by using a deep Hash mapping model; And searching a pre-constructed Hash code library for a target image whose Hamming distance difference with the to-be-retrieved Hash code satisfies a preset condition, outputting the target image as a retrieval result of the to-be-retrieved image in the image database, and mapping each image in the image database through a deep Hash mapping model to obtain the Hash code library. According to the image retrieval method and device, the problem that Hash codes of the same category of images are too consistent in the related art is effectivelysolved, and therefore accurate retrieval of the same category of images is achieved.

The invention discloses an image Hash coding method based on deep learning. The method includes following steps: step 1, adopting an image classification model GoogLeNet trained at an ImageNet image recognition database as an initialized basic network structure, and replacing the final classification layer of the GoogLeNet model by a Hash layer, wherein the unit number of the Hash layer is the bit number to be coded by images; step 2, optimizing parameters of the GoogLeNet model; and step 3, inputting images in an image retrieval data set to the optimized GoogLeNet model, and quantifying the floating point number output by the GoogLeNet model to binary codes to obtain the binary code of each image. According to method, combined optimization of image characteristics and a Hash function is realized, and the defect that the Hash codes obtained by learning of the conventional Hash method is not in accordance with the image characteristics is overcome.

The invention discloses a method for detecting an approximately repeated image. The method comprises: step one, symmetric hierarchy hash coding network feature extraction and description are carried out on to-be-matched images to obtain image features for constructing a locality sensitive hash index structure; step two, on the basis of the image features obtained at the step one, a locality sensitive hash index structure with a balanced load is constructed; and step three, a symmetric convolution neural network feature extraction is carried out on an inquired image and a corresponding hash code of the inquired image is calculated, correlation of the to-be-matched images are ranked based on the locality sensitive hash index structure with the balanced load constructed at the step two, and an image with the highest correlation is selected to be an approximately repeated image. Therefore, approximately repeated image detection can be carried out rapidly and effectively.

The invention relates to a model training method based on deep learning and hash coding. The method comprises the steps of regarding partially labeled image data as training data of a network model, and representing the training data as an analog-two-value hash code through a deep network, wherein the analogy-two-value hash code refers to a simulation two-value hash code of which the value is a continuous value; regarding the obtained analog-two-value hash code as input, connecting the analog-two-value hash code to one or more task layers, and using one or more tasks to conduct training at the same time; obtaining a two-value hash code which is used for representing the training data and carries feature information capable of being retrieved based on the analog-two-value hash code.

The invention discloses an attention mechanism and hash-based image retrieval method and device and a storage medium, and the method comprises the steps: obtaining the discrete hash code of each imagebased on a pre-trained attention mechanism-based hash feature extraction model for each image in an original data set, and building an image feature library; obtaining a discrete hash code of a to-be-detected image based on a pre-trained hash feature extraction model based on the attention mechanism; and querying data closest to the discrete hash code Hamming distance of the to-be-detected imagein an image feature library, wherein the image corresponding to the data is a retrieval result. The Hash feature extraction model based on the attention mechanism gives full play to the ability of thedeep convolutional neural network to extract the abstract semantic features of the image; and by embedding the attention module, the network can focus on the visual content capable of expressing semantic information of the whole image as much as possible in the image, the expression effect of hash coding is improved, and image retrieval is more accurate and faster.

The invention discloses a target tracking method based on triple twin hash network learning, and relates to the technical field of computer vision, target tracking and deep learning. According to themethod, firstly, a triple twin hash network is constructed, and the network is composed of a data input layer, a convolution feature extraction layer and a hash coding layer. In the initial training process of the network, a training data set and a random gradient descent back propagation algorithm are used for training the triple twinning Hash network, and after training is completed, the initialcapacity of target positioning can be obtained through the network. In the tracking process, firstly, an input image passes through a triple twin region recommendation network to obtain correspondingcandidate frames, then the candidate frames are input into a triple twin Hash network to be subjected to forward processing, the similarity between each candidate frame and a query sample is calculated, the candidate frame with the highest similarity is selected as a tracking target object, and therefore target tracking is achieved.

The invention relates to an image retrieval method for a multi-index disk Hash structure. The method comprises the following steps that in an offline index stage, high-dimensional features are extracted from media data; a Hash mapping method is adopted for mapping the high-dimensional features into Hash codes; the Hash codes are averagely partitioned into code substrings; the code substrings are inserted into corresponding B+ tree index structures based on a disk respectively; the steps are repeated m times, and the multi-index disk Hash structure is formed. In an on-line inquiry stage, the inquired media data is subjected to feature extraction, Hash coding and code partitioning, then the code substrings are obtained, r neighbor results of the codes in disk indexes are searched and combined till a k neighbor result is found, and the steps return. The advantages of the multi-index Hash method and the advantages of a B+ tree storage structure are combined, the retrieval accuracy and retrieval speed of the index structure are improved, and meanwhile the supported data volume is increased.

The invention provides a depth neural network training method and device based on a Hash coding layer. The method comprises the steps: obtaining a pre-trained neural network model which comprises a basic feature layer; Inserting a first full connection layer for Hash coding linear mapping and the Hash coding layer serving as an activation function behind the basic feature layer in the neural network model; inserting a second full connection layer serving as an output layer behind the Hash coding layer; and carrying out the training of the weight of the first full connection layer and the second full connection layer. According to the invention, the method takes the Hash coding layer as the activation function of depth neural network training, carries out the training of the neural network through employing a method of random gradient descent, achieves overall optimization, achieves the direct training of a model outputting a binary code, and does not need to relax the optimization conditions.

The invention discloses a rapid image retrieval method based on integrated Hash encoding, belongs to the technical field of digital image retrieval. The rapid image retrieval method comprises the following steps: firstly, extracting a training image and checking the SIFT characteristics of the image, and performing initial Hash encoding on the training image by using M Hash algorithms; subsequently relearning the initial Hash encoding result by using a consistency constraint rule in integrated learning, so as to obtain an integrated Hash mapping matrix; and finally performing integrated Hash encoding on the training image and the checking image again, and performing rapid retrieval by calculating the Hamming distance between the checking image and the training image on the basis of the integrated Hash encoding. According to the integrated Hash encoding used in the method, the characteristics and advantages of different Hash algorithms can be combined simultaneously, the problem that a single Hash algorithm is insufficient in judgment power and the application range is limited are solved, and thus rapid retrieval of images is relatively accurate and efficient.

The invention provides a method for rapidly filtering a key word. The method comprises the steps of: indexing the key word, and matching the key word. The method comprises the steps of: specifically carrying out hash encoding on each key word in a key word set, and carrying out key word matching by hash encoding, wherein the efficiency is higher than that by using a common multi-mode key word matching algorithm. An end mark is adopted to trigger key word matching, so as to achieve the rapid judgment that a current text can not be matched with any key word in the text scanning process; and the key word is matched through a hash value only when head and tail characters are consistent. Thus, the text scanning and matching speeds are greatly quickened.

The invention discloses an image classification and recognition method based on a twin network. According to the method, repeated inspection is carried out through Hash coding; preprocessing such as boundary frame prediction and affine transformation is simplified, and the data set quality is improved; Then, the test set and the training set are traversed through Hash coding, matched picture pairsand unmatched picture pairs are formed through combination in sequence, the matched picture pairs and the unmatched picture pairs are alternately input into a twin classification network for trainingfitting, and finally the classification effect that pictures of the same type are classified into the same type and different types can be effectively distinguished is achieved. According to the method, the defect of low prediction accuracy of an early-stage deep learning classification method when the test set is more than the training set and the category data is unbalanced is overcome, and theproblems that the classification data is unbalanced, the test set is more than the training set and the overall scale is small in an actual scene are solved. Besides, by encoding the picture data, the matching picture pair and the mismatching picture pair are analyzed, so that the accuracy of the twin classification network is improved, and a good example is provided for picture classification inan actual scene.

The invention discloses a book recommending method based on multi-view hash, comprising the following steps of 1) screening behavior data of a user on two views from a log collection system, wherein the behavior data comprise book clicking data and search data; 2) establishing user feature vectors of the user for clicking and searching the views; 3) obtaining a user hash code, a hash function and weights of the two views through a multi-view hash algorithm by utilization of the behavior data of the two views; 4) searching a similar user for a target user by utilization of the obtained user hash code; 5) obtaining a book set clicked by the similar user, taking the book set as a recommended candidate list, calculating the book preference degree of the target user, and returning front N books with maximal preference degree of the target user. According to the book recommending method based on the multi-view hash, on one hand, the behavior data of the user on two views can be integrated into the hash code, so that the book recommendation accuracy is improved; on the other hand, the Hamming distance calculation speed of the hash code is very fast, so that the book recommendation efficiency can be improved.

The invention provides a multi-label image retrieval method fusing triple loss and generative adversarial network, which comprises the following steps of: establishing a deep learning framework, and deploying a generative adversarial network model; inputting the image data set into the generative adversarial network model to obtain a multi-label image and triple data; constructing a triple loss function based on the multi-label image; selecting a first image from the image data set to train the deep hash coding network so as to obtain a trained deep hash coding network; selecting a preset number of second images from the image data set, and inputting the second images into the trained deep hash coding network to obtain a hash vector database; and inputting a first image to be retrieved into the trained deep hash coding network to retrieve a second image similar to the first image. According to the invention, the generative adversarial network is used to generate the multi-label generation picture similar to the data set sample, the training data volume is expanded, and the retrieval speed and precision of the image are improved.

The invention discloses a blocking perception Hash tracking method with shadow removing. The method comprises the steps of: determining shadow areas in an image according to the distribution characteristics of a shadow image in each channel grey-scale map of a CIELAB color space; then utilizing a color constancy theory to recover pixel points in the shadow areas to a non-shadow effect; combining blocking perception Hash coding values with color self-similarity for forming a similarity measure, and carrying out matching on tracked target sub-blocks of adjacent frames based on the similarity measure; and finally, combining the above sub-blocks to obtain the regional position of the tracked target in the current frame, and realizing the tracking of the tracked target in a video. The blocking perception Hash tracking method has the advantages that according to different moving ranges and deforming degrees of human body parts, a human body target is divided into eight sub-blocks, and on this basis, and the blocking perception Hash coding method is provided to solve the problem of an existing tracking algorithm that the tracking is unsuccessful when the human body is partially or totally shielded or partially rotated and when the illumination in the shadow areas and non-shadow areas of a natural scene changes suddenly.