54 results about "Invariant feature extraction" patented technology

The invention relates to a dynamic target position and attitude measurement method based on monocular vision at the tail end of a mechanical arm and belongs to the field of vision measurement. The method comprises the following steps: firstly calibrating with a video camera and calibrating with hands and eyes; then shooting two pictures with the video camera, extracting spatial feature points in target areas in the pictures by utilizing a scale-invariant feature extraction method and matching the feature points; resolving a fundamental matrix between the two pictures by utilizing an epipolar geometry constraint method to obtain an essential matrix, and further resolving a rotation transformation matrix and a displacement transformation matrix of the video camera; then performing three-dimensional reconstruction and scale correction on the feature points; and finally constructing a target coordinate system by utilizing the feature points after reconstruction so as to obtain the position and the attitude of a target relative to the video camera. According to the method provided by the invention, the monocular vision is adopted, the calculation process is simplified, the calibration with the hands and the eyes is used, and the elimination of error solutions in the measurement process of the position and the attitude of the video camera can be simplified. The method is suitable for measuring the relative positions and attitudes of stationery targets and low-dynamic targets.

The invention discloses a machine vision-based strawberry appearance quality judgment method. The method specifically comprises the following steps of 1: establishing a three-camera vision platform and then all-directionally acquiring strawberry RGB images in an industrial scene; 2: performing image segmentation by utilizing the acquired images in the step 1, then defining fruit regions of interest, and extracting strawberry fruit contours; 3: extracting shape feature vectors of strawberries by using a shape invariant feature extraction operator through utilizing the extracted strawberry fruit contours in the step 2, thereby overcoming the problems of different sizes, positions and orientations of the strawberries; and 4: training a support vector machine by utilizing the acquired sample images of strawberries in different shapes in the step 1 and the extracted feature vectors in the steps 2 and 3, and determining optimal parameters and classification accuracy of the support vector machine. The invention provides a strawberry appearance quality judgment method with the characteristics of high accuracy, robustness, real-time property and no damage for the problems of low manual identification efficiency, high mechanical damage rate and the like, and actual usage requirements are met.

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 relates to a profiled fiber automatic recognition method based on scale invariant feature extraction and vector machine classification. The method is characterized by comprising the following steps: (1) inputting the cross-section micro-sample images of a profiled fiber to be recognized and pre-processing the images; (2) carrying out binary processing of the pre-processed images to obtain a binary image of the fiber; (3) extracting the fiber object in the binary image by a separation algorithm to obtain the independent image of each fiber in the binary image; (4) extracting the eigenvector of the existing profiled fiber by an SIFT (scale invariant feature transform) algorithm to serve as the input, and training by an SVM (support vector machine) classification algorithm; and (5) using the independent image obtained in the step (3) as the input to determine the type of profiled fiber, to which the fiber on each independent image belongs, by the classification algorithm trained in the step (4). By utilizing the method, automatic classification and recognition of profiled fibers in complex background, tilting fibers, deformed fibers, broken fibers, fibers in illumination changing environment and the like can be achieved, and the problem of poor effectiveness of image feature extraction is solved.

The invention discloses a frequency displacement invariant feature extracting method considering multifactor in voice, and the method is used for distinguishing text-independent speakers under a complex environment. The method comprises the step of: in consideration of the time, frequency, scale and phase information of voice, performing multifactor characterization on a voice signal energy spectrum through two-dimensional plurality wavelet transform of different scales and phases, in consideration of the displacement invariant feature of the frequency, calculating a displacement invariant feature projection matrix on a frequency order through a convolution-type non-negative tensor decomposition method to obtain a multifactor sparse feature, decorrelating the feature through discrete cosine transform, and calculating the first order and second order difference coefficients of the feature, thus finally obtaining speaker feature with good robustness. The frequency displacement invariant feature extracting method considering multifactor in voice disclosed by the invention calculates the multifactor frequency displacement invariant feature with robustness through the convolution-type non-negative tensor decomposition method to distinguish the text-independent speakers under a noise environment, so that good distinguishing accuracy is good.

The invention discloses an illumination normalization method for processing face images. The illumination normalization method includes steps of reading face images to be processed and taking logarithm of the face images to be processed; computing images of a shadow layer; computing images of a reflecting layer and performing exponential transformation; selecting sample images and computing a histogram of the sample images; normalizing the images of the reflecting layer by a histogram matching method to obtain the images of the to-be-processed face images corrected by the illumination normalization method. By an edge-preserving filter, feature information on a large-scale layer can be kept. By adopting the sample images in the optimum vision area, mean value and variance of the corrected images of the reflecting layer can be kept in the optimum vision area, a subsequent face identification system is facilitated, accuracy in face identification can be improved, and the problem that the feature information is greatly lost to cause disadvantages to subsequent face identification during face identification in existing illumination invariant feature extracting method is solved.

The invention discloses a device and extraction method for extracting invariant characteristics of the local rotation of an image. The device is characterized by comprising a characteristic point positioning unit, a neighborhood dividing unit and a histogram statistical unit, wherein the characteristic point positioning unit is used for positioning a characteristic point in the image; the neighborhood dividing unit is used for dividing the neighborhood of the characteristic point into a plurality of concentric annular areas and further dividing the annular areas into a plurality of sectorial areas; and the histogram statistical unit is used for performing statistics of local binary characteristics in the sectorial areas respectively so as to establish a histogram. As the means for extracting the invariant characteristics of the local rotation is adopted, the computational complexity is greatly reduced, less memory space is occupied, therefore the speed of a computer for processing the image is improved under the condition that the positioning accuracy is ensured.

The invention discloses a speech recognition method based on a domain-invariant feature, and the method applies a speech domain-invariant feature extraction model to an end-to-end speech recognition model. The feature extraction model used in the speech recognition method aims at the robustness problem, and by adding more types of speech data to train the speech feature extraction model, better parameters can be obtained and a better domain-invariant feature extraction model can be obtained. The speech recognition method based on the domain-invariant feature uses unlabeled pure speech data totrain the feature extraction model, uses a small number of speeches with text annotation to train an end-to-end acoustic model, and provides important technical support for improving the robustness ofthe end-to-end acoustic model. Compared with the prior art, the speech recognition method based on the domain-invariant feature has higher recognition accuracy in different noise environments, smaller task quantity of speech annotation tasks, and faster training and testing speed of the models.

The invention discloses a method for describing a local characteristic of an image, which is used for describing characteristic manifold of a characteristic point extracted through a size-invariable feature extraction method. The method is characterized in that: a series of affine transformation is processed for a local area of the characteristic point, a size-invariable characteristic description vector of the characteristic point is correspondently extracted from each variable image to form a characteristic vector set, and the characteristic vector set is further simulated through a linear subspace set to be used as a characteristic descriptor of the characteristic manifold. The characteristic description method comprises the following steps that: the local area image of the characteristic point is extracted, a series of transformation is processed for the local image, the size-invariable characteristic description is extracted for the characteristic point of each variable image, the characteristic vector set is formed, the linear subspace is adopted to approach the characteristic vector set, and the characteristic descriptor is generated. The input of the method is a series of characteristic points which are expressed by coordinate positions, and the output of the method is the characteristic description expressed by a plurality of linear subspaces.

The invention relates to an image invariant feature extraction method based on the circular trace transform, and belongs to the field of image processing and computer vision. In this method, a circular trace line of the new circular trace transform is used to track an image, functional results on the circular trace line are mapped to a space generated by the three parameters of radius, length and angle, the result of the circular trace transform is obtained, the result is further subjected to functional integration, and image quadruple circular trace spatial features can be obtained. Different circular trace transform results can be obtained by using different functions on the circular trace line, that is, different texture properties of the image can be represented and multi-dimensional circular trace transform texture feature information is generated. According to the invention, the texture features extracted by the circular trace transform have an RST invariant property, deeper image information can be obtained, a better resolution capability is gained for images containing circular and arc-shaped textures, and the recognition capability is improved significantly in the case of fewer training samples.

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 provides a real-time target matching tracking method and a system. According to the method, feature information of pixels contained in each of at least two images which are taken in the same place at similar shooting time is analyzed, so that at least one target area from each image is acquired; and then matching target areas of each target area in the corresponding image according to area information occupied by the target area in the corresponding image are determined, wherein the matching target areas are positioned in other images. The areas, which is occupied by a target, in the images are used as tracking objects, the matching target areas, which are positioned in the other images, of each target area are determined based on the area information occupied by the target area in the corresponding image, existing invariant feature extraction algorithm is omitted, accordingly, compared with the prior art, the method is low in complexity, real-time tracking to the target areas can be realized easily, and the method can be applied to more monitoring places.

The invention discloses a nondestructive testing method for agricultural product quality indicators. The nondestructive testing method comprises the following steps: Step 1, building a machine visiondetecting platform, and then comprehensively collecting RGB images of agricultural products in a visual sensor; Step 2, carrying out image segmentation on the images collected in Step 1, then dividingregions of agricultural products with interest, and extracting the contours of fruits of the agricultural products; Step 3, with the contours of the fruits of the agricultural products extracted in Step 2, extracting a shape feature vector with a shape invariant feature extraction operator, so as to overcome the problem of different sizes, positions and orientations of the agricultural products;and Step 4, with the images of the agricultural products of different shapes collected in the Step 1 and the feature vector extracted in Steps 2 and 3, training a support vector machine to determine the optimal parameters and the detection accuracy of the support vector machine. According to the invention, the machine vision technology is adopted to realize the non-destructive detection of the quality of agricultural products, and the nondestructive testing method has a very high real-time property, is reasonable and has high detection accuracy.

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 invention provides a method and a device for extracting scale-invariant features of images in video streams and a method and a device for recognizing objects on the basis of the method for extracting the features. The images comprise gray images and corresponding parallax images, correspond to moments t1 in the aspect of time, and include precedent images and posterior images in the video streams in the aspect of time. The method for extracting the features can include positioning critical points in the images; generating description regions around the critical points, and describing each description region around the corresponding critical point in four dimensions x, y, z and t; generating descriptors for each critical point on the basis of the description region of the critical point. Values of each description region in the range of each of the corresponding dimensions x, y, z and t are not equal to zero. The descriptors are used as the scale-invariant features of the critical points. The methods and the devices have the advantages that information of time domains, information of depth domains and information of image planes are closely combined with one another, so that the four-dimensional scale-invariant features can be extracted, and the methods and the devices are suitable to be applied to machine learning.

The invention discloses an illumination invariant face feature extraction method by means of logarithmic transformation and smoothing filtering. The illumination invariant face feature extraction method comprises following steps of transforming a face image to a log domain; processing the face image by means of smoothing filtering; subtracting an original image from the processed image to eliminate the illumination component, and obtaining the result which is the illumination invariant face feature which needs to be extracted. The illumination invariant face feature extraction method has certain inhibiting effect on illumination from different angles, and the recognition rate of the face image processed by the above method is higher than that of the face image processed by an existing algorithm. According to the invention, the face image is subjected to illumination processing by means of logarithmic transformation and smoothing filtering, the method is simple, and the recognition rate is fast.

The invention provides a ship retrieval method and device based on an SIFT algorithm, and the method comprises the steps: obtaining a to-be-detected image and a target ship image, and extracting a ship target region image from the to-be-detected image; obtaining key point feature vectors of the ship target area image and the target ship image according to an SIFT algorithm; determining the similarity between the ship target area image and the target ship image according to the key point feature vector; and when the similarity is higher than a first threshold, returning the ship target area image as a detection result. According to the invention, the SIFT algorithm is adopted to obtain a ship target area image and a key point feature vector in the target ship image; and the similarity between the ship target area image and the target ship image is calcculated according to the key point feature vector, and the ship target area image of which the similarity exceeds a first threshold valueis returned to serve as a detection result, so that invariant feature extraction can be realized, and the target ship image can be accurately detected from numerous ships.

The invention discloses an open set domain adaptive image classification method and system in the technical field of image recognition, and the method and system enable a network to better obtain domain invariant features through a channel attention mechanism, facilitate the migration of the features, and enable the features to be trained more easily. The method comprises the following steps: respectively inputting a labeled sample obtained from a source domain and an unlabeled sample obtained from a target domain into a feature extractor based on a channel attention module to obtain a weighted multi-channel feature map; sending the weighted multi-channel feature map into a label classifier, dividing samples with labels into K known categories, and dividing samples without labels into K known categories visible in a source domain and an unknown category invisible in the source domain; sending known categories from a source domain and a target domain to a domain discriminator, and enhancing domain invariant feature extraction based on the generative adversarial network; reducing inter-domain differences between the source domain and the target domain based on covariance matching.

The invention provides an unsupervised domain adaptation method and system based on target domain self-supervised learning. The unsupervised domain adaptation method comprises a domain invariant feature extraction step, an image feature stepped domain alignment step, a semantic segmentation step, an edge generation step, a segmentation image domain alignment step, an edge image domain alignment step and an edge consistency constraint step. According to the method, effective self-supervised learning is carried out on the target field, so that the unsupervised segmentation precision of the target field is improved and good field adaptation is realized.

The invention discloses a pedestrian re-identification method based on human body posture invariant features, which comprises the following steps of: (10) original image preprocessing: preprocessing an original pedestrian image by utilizing a multi-scale retina enhancement algorithm with color recovery factors to obtain a corrected image with rich local color information; (20) model training: taking the original pedestrian image and the correction image as input, and training a ResNet 50-based pseudo twin neural network model; (30) invariant feature extraction: extracting human body posture invariant features by using the trained pseudo-twin neural network model; and (40) pedestrian re-identification: re-identifying the pedestrian according to the human body posture invariant feature. According to the pedestrian re-identification method based on the human body posture invariant features, the identification effect is less influenced by illumination conditions and human body posture changes.

Described is a system for decoding and validating memory consolidation. During operation, the system receives electroencephalographic (EEG) data while a subject is performing a specific task. Nuisance signals are then removed from the EEG data, resulting in a nuisance free signal. Skill feature vectors are generated from the nuisance free signal using time-invariant feature extraction. A skill classifier can then be trained for the specific task based on the skill feature vectors to generate a subject specific model regarding a memory replay for the specific task. Finally, electrodes in a neural cap are activated based on the memory replay.

The invention discloses a method for extracting a face illumination invariant feature based on smoothing filtering. The method includes the steps of processing a face image using smoothing filtering; and dividing the processed image by a pixel corresponding to the original image, eliminating a face illumination part that slowly changes and obtaining the face illumination invariant feature that needs to be extracted. The invention has a certain inhibitory effect on the illumination of different angles, and the recognition rate of the processed face image is higher than that of the existing algorithm. According to the invention, the illumination processing is performed for the face image by the smoothing filtering and the division operation, the method is simple and the recognition speed is high.

The invention discloses a face attribute invariant robustness face recognition method and system. The method comprises the following steps: S1, extracting face features; s2, estimating human face attributes; s3, constructing and selecting a face attribute sub-dictionary; s4, carrying out attribute-invariant face feature extraction; s5, carrying out face classification and recognition; in the system, a face depth feature extraction module is connected with a face attribute recognition module and a face attribute dictionary construction module; the face attribute recognition module is sequentially connected with the face attribute sub-dictionary selection module, the face attribute invariant feature extraction module and the face classification recognition module, and the face attribute sub-dictionary selection module is connected with the face attribute sub-dictionary selection module through the attribute over-complete dictionary. The method and system can solve the problem of face recognition in complex environments such as simultaneous changes of illumination, postures, expressions, shielding and resolution of the face and combination changes of the changes, keeps face attributesunchanged, and are high in robustness.

The invention discloses a face illumination invariant feature extraction method based on a Kirsch operator. The method comprises the following steps that the Kirsch operator is used to sharpen a face image, and detail features of the face are extracted; and the sharpened image is divided by a corresponding pixel of the original image to eliminate a slowly-change face illumination part and obtain a face illumination invariant feature which needs extracting. According to the method of the invention, illumination of different angles can be inhibited to certain extent, and the identification rate of the processed face image is higher than the face identification rate processed by a present algorithm; illumination processing is carried out on the face image by utilizing the Kirsch operator and the division, the method is simple and the identification speed is high.

The invention discloses a frequency displacement invariant feature extracting method considering multifactor in voice, and the method is used for distinguishing text-independent speakers under a complex environment. The method comprises the step of: in consideration of the time, frequency, scale and phase information of voice, performing multifactor characterization on a voice signal energy spectrum through two-dimensional plurality wavelet transform of different scales and phases, in consideration of the displacement invariant feature of the frequency, calculating a displacement invariant feature projection matrix on a frequency order through a convolution-type non-negative tensor decomposition method to obtain a multifactor sparse feature, decorrelating the feature through discrete cosine transform, and calculating the first order and second order difference coefficients of the feature, thus finally obtaining speaker feature with good robustness. The frequency displacement invariant feature extracting method considering multifactor in voice disclosed by the invention calculates the multifactor frequency displacement invariant feature with robustness through the convolution-type non-negative tensor decomposition method to distinguish the text-independent speakers under a noise environment, so that good distinguishing accuracy is good.

The invention discloses a method for extracting human face illumination invariant features by utilizing a Wallis operator. The method comprises the following steps of: transforming an original image to a logarithm domain; and sharpening a human face image by utilizing the Wallis operator, extracting human face detail features, and eliminating human face illumination components at the same time, wherein the obtained result is a to-be-extracted human face illumination invariant feature. The method has a certain inhibiting effect for illuminations at different angles; the identification rate of the processed human face images is higher than the human face identification rate obtained through existing method; and according to the method, light processing is carried out on human face images by adoption of logarithm transformation and the Wallis operation, so that the method is simple and high in identification speed.

The invention provides a migratable motor fault diagnosis method based on a depth auto-encoder and domain invariant feature extraction, and the method comprises the steps: carrying out the signal processing and mixed domain feature extraction of a motor vibration signal through dual-tree complex wavelet packet transformation; secondly, providing a domain invariant feature selection method based on importance scores and inter-domain difference measurement, and selecting features with high fault resolution capability and domain invariant characteristics; then, pre-training a depth automatic encoder (a source model) by adopting the selected domain invariant features so as to enhance the fault characterization capability of the depth features; and migrating the parameters of the source model to the target model with the same structure, and finely adjusting the target model by using the normal state feature data of the target domain. And finally, classifying target domain faults by adopting the fine-tuned target model. According to the method provided by the invention, the cross-domain fault diagnosis accuracy can be obviously improved, and the method has stronger availability, stability and advantages in an actual industrial scene.

The invention discloses face illumination invariant feature extraction method by utilizing logarithm transformation and a Laplacian operator. The method comprises the following steps that a face image is transformed to a log domain; and the Laplacian operator is used to sharpen the face image, detail features of the face are extracted, a face illumination component is eliminated, and an obtained result refers to a face illumination invariant feature which needs extracting. According to the method of the invention, illumination of different angles can be inhibited to certain extent, and the identification rate of the processed face image is higher than the face identification rate processed by a present algorithm; illumination processing is carried out on the face image by utilizing the logarithm transformation and the Laplacian operator, the method is simple and the identification speed is high.

The invention discloses a face illumination invariant feature extraction method based on a Laplacian operator, and belongs to the technical field of mode identification. The method comprises the following steps that the Laplacian operator is used to sharpen a face image, and detail features of the face are extracted; and the sharpened image is divided by a corresponding pixel of the original image to eliminate a slowly-change face illumination part and obtain a face illumination invariant feature which needs extracting. According to the method of the invention, illumination of different angles can be inhibited to certain extent, and the identification rate of the processed face image is higher than the face identification rate processed by a present algorithm; illumination processing is carried out on the face image by utilizing the Laplacian operator and the division, the method is simple and the identification speed is high.

The invention discloses a face illumination invariant feature extraction method by using logarithmic transformation and a Roberts operator. The method comprises the following steps: a face image is transformed into a logarithmic domain; the Roberts operator is used to sharpen the face image, the details of the face are extracted, an illumination component is eliminated, and the obtained result is the face illumination invariant feature which needs to be extracted. A certain suppression role is played to illumination of different angles, and the face image recognition rate after processing is higher than that after processing in the existing method; and the logarithmic transformation and the Roberts operator are adopted to carry out illumination processing on the face image, the method is simple, and the recognition speed is quick.