155 results about "Translation invariance" patented technology

The invention discloses a building overall-deformation monitoring method based on a three-dimensional laser scanning technology. The method includes the following steps: obtaining building point cloud data, segmenting the point cloud data, regularizing the data, performing deformation analysis on a point cloud model and judging a building overall-deformation condition. The building overall-deformation monitoring method based on the three-dimensional laser scanning technology is capable of rapidly extracting three-dimensional space characteristics of the point cloud model from the building mass point cloud data and the space characteristics have rotation and translation invariance so that attitude of the building in a three-dimensional space can be expressed uniquely and according to the variance condition of the three-dimensional space characteristics of the building point cloud model, the overall deformation condition of the building, such as sedimentation, inclination and rotation and the like in the three-dimensional space can be analyzed and judged. The building overall-deformation monitoring method based on the three-dimensional laser scanning technology makes full use of complete point cloud data of the surface of the building to analyze the overall deformation condition of the building and the measurement and analysis technology based on faces effectively prevents locality and one-sidedness brought by a deformation analysis result in traditional deformation monitoring based on points.

The invention provides an image recognition method, device and system, and relates to the technical field of image recognition. The method includes the step of acquiring an image to be recognized; inputting the image to be recognized into a convolution neural network model, wherein the convolution neural network model includes a translation invariance pyramid residual module, a hourglass module and a cascaded pyramid network; extracting the features of the image to be recognized step by step by the translation invariance pyramid residual module, the hourglass module and the cascaded pyramid network of the convolution neural network model to obtain the recognition result of the image to be recognized, wherein the identification result includes a position of at least one target point. The invention can recognize based on the image taken by the monocular camera, and can improve the accuracy of image recognition.

The invention relates to an object segmentation method based on a priori shape and a CV (Computer Vision) model. The method comprises the following steps of: constructing a signed distance function by selecting the priori shape; carrying out affine transformation on the signed distance function of the priori shape according to vectors of affine transformation parameters, wherein the affine transformation parameters are changed and finally tend to stabilize in the process of iterating a level set function; and finally, obtaining various affine transformation parameters of the next time according to a level set iteration formula evolution movable outline curved line and various affine parameter iteration formulae. According to the method disclosed by the invention, on the basis of keeping rotary, zoom and translation invariance of a priori shape model, stretching in X and Y directions and shear unchanged constraint energy item can be increased; and a space object with larger posture transformation under a complex background can be well segmented by expanding the self-adaptive transformation of the priori shape.

The invention discloses a method for measuring a relative pose of a noncooperative target. The method comprises the steps: selecting an object to be measured with equilateral triangle characteristics as the noncooperative target, and establishing a single line structured light based visual measurement system model; extracting three straight lines of the object to be measured and crossing points between single line structured light and the object to be measured, so as to obtain coordinates of five characteristic points of the object to be measured in an image coordinate system; carrying out calculation according to geometric constraint conditions of the object to be measured, so as to obtain coordinates of the five characteristic points of the object to be measured in a camera coordinate system; carrying out calculation according to distance translation invariance and the geometric constraint conditions of the object to be measured so as to obtain coordinates of the five characteristic points of the object to be measured in a world coordinate system, substituting the coordinates into a transformation equation of the camera coordinate system and the world coordinate system, and carrying out solving, so as to obtain the relative pose of the object to be measured, i.e., the relative pose of the noncooperative target. According to the method, through increasing an auxiliary line structured light source to monocular vision, the measurement on the relative pose of triangular characteristics of unknown dimensions is achieved.

The invention relates to a recognition method and a recognition device based on biological characteristics. According to the recognition method, wavelet packet multilayer decomposition characteristic parameters are used for generating new frequency characteristic quantities; due to the fact that wavelet coefficients have a large difference, energy of signals at all decomposition nodes not only can reflect energy distribution of the signals in all frequency bands, but also has translation invariance, as a result, the frequency characteristic quantities can highly reflect user characteristics and improve the recognition rate of user signatures; and further, fingerprint characteristics and handwritten signature characteristics are combined, fingerprint information is collected while a user gives a signature, and therefore a high recognition rate of fingerprints is used for making up for a low high recognition rate of handwritten signatures. Accordingly, a handwriting pen provided with a fingerprint collection module at the pen holding end in an inlaid mode is used, and therefore the fingerprint information of the user can be collected while the user gives the signature, namely two types of information can be collected at the time of one action. A biometric feature recognition system which combines the fingerprints and the handwritten signatures and is studied out according to the thinking is high in recognition rate and acceptability.

The invention discloses a robust digital image adaptive zero-watermarking method. The method mainly comprises two parts, namely zero-watermark embedding and zero-watermark detection, wherein both the zero-watermark embedding and the zero-watermark detection are performed in a composite domain of discrete wavelet transform and discrete fourier transform; the characteristics of high stability of a wavelet approximation subgraph obtained by the discrete wavelet transform and translation invariance of an amplitude spectrum obtained by the discrete fourier transform are fully utilized, so that the method has high robustness, can resist common image processing and is completely immune from image translation attack; meanwhile, because a binary digital watermark is not embedded into the original digital image but is registered into a watermark database, no damage is caused to the original digital image data, the problem of quality reduction of the image does not exist, the embedded binary digital watermark is completely imperceptible, and the contradiction between the robustness and the imperceptibleness of the digital watermark can be balanced well.

The invention discloses a shape matching and target recognition method based on a PCA-SC algorithm. The method comprises the steps of carrying out preprocessing on a target image, filtering part of noises in the target image, extracting the edge of the target image, extracting information of boundary contour points, working out the rectangular coordinate parameters of the contour points, converting the contour points from rectangular coordinates into polar coordinates, obtaining a corresponding logarithmic polar histogram of each point to forming a local feature descriptor, forming a covariance matrix, extracting a corresponding feature vector of a larger characteristic value of the matrix, adopting a linear transformation method to drop the matrix from high dimension to low dimension, forming a new characteristic matrix, wherein the new characteristic matrix is used for the shape matching and the target recognition, calculating matching degree, and obtaining a matching degree value between the target image and each template image. According to the shape matching and target recognition method based on the PCA-SC algorithm, characteristic extracting and effective representation for the image can be achieved, scale invariance, rotation invariance and translation invariance are achieved, accuracy rate and efficiency are improved, and interference of the noise is effectively restrained.

The invention provides a fine granularity classification recognition method and an object part location and feature extraction method thereof. The fine granularity classification recognition method and the object part location and feature extraction method thereof well achieve object part location and feature expression in fine granularity classification recognition. For object part location, a series of part detectors trained by supervised learning are utilized, the methods just detect the part with small deformation in consideration of the posture change and deformation influence of targets to be located, different detectors are trained for the same object part by adopting the posture clustering method, and therefore the posture change of objects is taken into account. For feature expression of the objects or parts, features are extracted at multiple dimensions and multiple positions according to the methods and then fused to be used for final object expression, and therefore the features have certain dimension and translation invariance. According to the methods, object part location and feature expression have certain complementarity at the same time, and therefore the accuracy of fine granularity classification recognition can be effectively improved.

The invention provides an image fusion method and a device based on NSST and adaptive dual-channel PCNN. NSST is used as a multi-scale transform tool to decompose the image, NSST removes the sample operation in the decomposition phase, and has translation invariance, at the same time, a localize small size shearing filter can avoid spectral aliasing, so that the effect of image decomposition and reconstruction is better, using an adaptive simplified two-channel PCNN model with three kinds of link strengths, through three different link strengths, the features of the image are integrated, the infrared and visible images are fused effectively, the contrast is high, the texture information and detail information are well preserved, and the image transition is natural, no artifact phenomenon occurs.

The invention discloses a feature extraction and dimension-reduced neural network-based visual SLAM closed-loop detection method. According to the feature extraction and dimension-reduced neural network-based visual SLAM closed-loop detection method, a convolutional neural network model is trained through a large number of data set to endow a network with feature learning capacity, so that similarity comparison between images can be converted into similarity comparison between feature vectors; for further improving the detecting speed, the last layer of the convolutional neural network is provided with an auto-encoder network to dimension-reducing extracted image features; the convolutional neural network has the advantages of translation invariance, scale invariance and the like and accordingly can effectively overcome the shortcoming that traditional artificial feature extraction is sensitive to environmental change and achieve a higher feature extraction speed. The feature extraction and dimension-reduced neural network-based visual SLAM closed-loop detection method can overcome the shortcomings of short feature extraction time and large influence of environmental change and light change in traditional visual SLAM closed-loop detection methods, effectively improve the accuracy and recall rate of closed-loop detection and achieve high significance to structuring globally uniform environmental maps.

The invention provides a massive image classification method based on a deep vector of locally aggregated descriptors (VLAD). The method is characterized in that through multi-layered image VLAD extraction, more abstract and richer image expression is finally obtained, so that the purpose of higher image classification accuracy is achieved. The method provided by the invention has the advantages that an adopted deep model can capture more abstract and richer feature information from an image during stepwise feature extraction, and average pooling operation is further conducted, so that captured image features achieve translation invariance, which is critical for the increasing image quantity and the image variety.

The invention discloses a high-resolution rapid deconvolution sound source imaging algorithm which is characterized by comprising the following steps: calculating a point spread function of a sound source at the central position of a sound source calculation plane by utilizing approximation space translation invariance of the point spread function in the process of constructing a point spread function matrix of the deconvolution sound source imaging algorithm; constructing the point spread function matrix through a method for circularly shifting the point spread function at the central position upwards and downwards, so that calculation of the total point spread functions is avoided, the calculated amount, and the calculation speed and efficiency are improved; realizing rapid sparse deconvolution reconstruction of sound source intensity energy distribution through an orthogonal matching pursuit algorithm by utilizing space sparse prior of the sound source and combining compressed sensing in the deconvolution reconstruction process of the sound source intensity energy distribution, so that the iterations are reduced, and the calculation efficiency and resolution ratio are improved. The algorithm disclosed by the invention has high calculation efficiency and spatial resolution and is capable of well rapidly identifying and locating the position of the sound source in the space.

The present invention discloses a cement notch groove pavement image noise reduction enhancement and crack feature extraction method aiming at the problems that the pavement contrast is too low causedby external factors and the payment spots and notch groove autointerference caused by pavement materials. The method comprises the following steps of: employing an improved local adaptive contrast enhancement algorithm to enhance image contrast after graying processing of an original cement pavement image; employing translation invariance Shearlet transform denoising algorithm of an improved P-Mmodel to remove speckle noise caused by pavement materials; employing a cement notch groove pavement image smoothing model established based on an unidirectional total variation UTV model to the imageafter denoising to remove a pavement notch groove influencing feature extraction; and combining a connected domain mark method, a projection method and a rectangular frame method to extract a crack type determination method and a crack feature calculation method to achieve digital description of crack features. The cement notch groove pavement image noise reduction enhancement and crack feature extraction method is systematic and comprehensive, small in calculated amount and easy to apply.

The invention provides an improved end-to-end speech recognition method, which combines a convolutional neural network and a self-attention mechanism and utilizes a CTC training criterion to train anend-to-end speech recognition model. The model is mainly composed of three parts: (1) a depth two-dimensional convolution part; (2) a self-attention part; (3) a full connection layer. The first part of the model effectively extracts features of a time axis and a frequency domain axis of a voice signal through two-dimensional convolution, and translation invariance is achieved; the second part enables the voice signal to be fully combined with context through the self-attention mechanism; the third part classifies features of each frame of voice; and finally model parameters are updated throughthe CTC training criterion. According to the model, the self-attention mechanism is innovatively added into a neural network-CTC framework, so that end-to-end speech recognition is realized, and therecognition effect is improved.

The invention relates to a DWT-SVD geometric attack prevention blind watermark method based on SIFT. The method comprises a watermark embedding method and a watermark extraction method. The watermark embedding method includes the steps that discrete waterlet transformation is conducted on an original image, a low-frequency sub-band of the original image is divided into small blocks, singular value decomposition is conducted on each small block, chaotic encryption is conducted on watermarks to be embedded, and the watermarks are embedded into the maximum singular value of each small block through an optimal quantitative method, the SIFT feature points of a watermark image are stored as secret keys, the types of geometric attacks are judged according to the secret keys, and correction is conducted. The watermark extraction method is the inverse process of the watermark embedding method and includes the steps of correction of attacked images, watermark extraction and watermark decryption and restoration. According to the method, by the utilization of the rotation, zoom and translation invariance of the SIFT feature points and the combination of the advantages of DWT and SVD in the digital watermark aspect, robustness on geometric attacks is remarkably improved.

The present invention relates to a contour chord angle feature based identification method for a blocked target. The method comprises: establishing a template library of local features of a plurality of target images; extracting a contour feature of a target edge; constructing a chord angle feature descriptor of each contour point; describing a blocked contour by using a self-containing attribute of the chord angle feature descriptor, to obtain a chord angle feature description matrix of contour segments; calculating distances between the chord angle feature descriptors of target image contour points and the chord angle feature descriptors of the contour points of the local feature in the template library by using an L1 measurement method, to obtain a matching cost matrix; and calculating a similarity of the matching cost matrix by using an integral graph algorithm, so as to identify a partially blocked object. According to the present invention, a contour space position feature of a target shape can be extracted, and blocked targets can be identified, and scale invariance, rotational invariance and translation invariance are achieved, thereby increasing the accuracy and robustness of target identification and shape search.

The invention discloses an image retrieval method based on a weight color-sift characteristic dictionary. The image retrieval method based on the weight color-sift characteristic dictionary comprises the following steps that training images are selected in images to be retrieved randomly, the edges of the training images are extracted, the color-sift characteristics of edge points of all the training images are extracted, and a characteristic dictionary is constructed according to the color-sift characteristics, an image which needs to be retrieved is input and the color-sift characteristics of the retrieval image and the edge points of the image to be retrieved are extracted, and weight histogram characteristics of the retrieval image and the image to be retrieved are extracted based on the characteristic dictionary; similarity matching based on the weight histogram characteristics is conducted on the retrieval image and the images to be retrieved in the database based on the weight histogram characteristics; whether all the images to be retrieved in the database are all traversed is detected,, the result is matched according to similarity and the image searching result is displayed if all the images to be retrieved in the database are all traversed, and the similarity matching is conducted again if all the images to be retrieved in the database are not traversed. The image retrieval method based on the weight color-sift characteristic dictionary improves accuracy and callback rate during searching of a large-scaled image database, has dimension invariance, translation invariance and rotation invariance, and has the advantages that the image retrieval method based on the color-sift characteristic dictionary has locality, particularity, multi-amount property and high-efficiency property.

The invention discloses a malicious code classification method based on a convolutional neural network. According to the method, malicious codes are mapped into single-channel signals, then a spectrogram of the signals is generated according to a signal processing method, an image scaling algorithm is used to convert the spectrogram into a grayscale map of a constant size, and finally the convolutional neural network is used to realize classification of the malicious codes. Through the method, the corresponding spectrogram is generated after the malicious codes are mapped into the signal-channel signals, sufficient context information of the malicious codes can be obtained, and the context information not only reflects time domain and frequency domain information of the signals but also can reflect local and global information of the signals; and besides, thanks to local translation invariance and other characteristics of the convolutional neural network, substantive characteristics ofthe malicious codes can be well obtained, code rearrangement, garbage code insertion and other situations are effectively overcome, and classification precision of the malicious codes is improved.

The invention discloses a hybrid transform domain image zero-watermarking method based on variable parameter chaotic mapping. The method comprises a zero watermark embedding process and a zero watermark detection process. The embedding and detection of the zero watermark are both completed in a non-subsampled pyramid decomposition and discrete cosine transform hybrid transform domain. The translation invariance characteristic of non-subsampled pyramid decomposition of the image and the stability of the alternating current coefficient symbol at the special position in each coefficient block after the image is subjected to block discrete cosine transform are fully utilized, so that the method has relatively good robustness. Experimental results show that the method has good robustness for resisting various common image processing attacks, such as noise addition, filtering, JPEG compression, scaling, rotation attacks at any angle and the like, and has effectiveness and certain practical value.

The present invention discloses a method and device for gesture identification based on substantial feature point extraction. The device comprises: an extraction module configured to obtain shape of a gesture to be identified, extract an unclosed contour from the edges of the shape of the gesture to be identified and obtain coordinates of all the contour points on the contour; a calculation module configured to calculate the area parameters of each contour point, perform screening of the contour points according to the area parameters, extract the substantial feature points and take the area parameters of a substantial feature point sequence and the point sequence parameters after normalization as the feature parameters of the contour; and a matching module configured to facilitate the feature parameters of the substantial feature points, perform matching of the gestures to be identified and templates in a preset template library, obtain the optimal matching template of the gesture to be identified and determine the type of the optimal matching template as the type of the gesture to be identified. The method and device for gesture identification based on the substantial feature point extraction have good performances such as translation invariance, rotation invariance, scale invariance and hinging invariance while effectively extracting and expressing gesture shape features so as to effectively inhibit noise interference.

The invention discloses a robust digital audio watermark embedding and detection method based on polar harmonic transform. Advantages of great frequency domain localization features, multi-scale, redundancy and translation invariance, the features of non-stationary of greatly describing signals and maintaining the shape of the signals and the same length of audio frequency of each sub-band after decomposition with that of the original audio frequency of nonsubsampled wavelet transform are utilized; then great numerical stability of polar harmonic transform is also utilized, and watermarks are embedded in stable torque values to resist the normal signal attack; and then the synchronization code technology is combined. The experimental results indicate that normal signal processing can be resisted and the desynchronization attack can also be effectively resisted so that the method has great imperceptibility and robustness. Besides, the method also has advantages of being simple in design and easy to realize.

Disclosed is an obtaining method of measurement matrixes of a compression spectral imaging system based on random wave front phase modulation. When the method is used for obtaining the measurement matrixes of the compression spectral imaging system based on random wave front phase modulation, a broadband hot light source, a spectral filtering device, a pinhole diaphragm and a magnification imaging lens need to be added in a light path of the system. According to the obtaining method, the translation invariance of the local space of the compression spectral imaging system based on the random wave front phase modulation is used, and the measurement matrixes of the system are obtained through low measuring frequency; a high-contrast-ratio speckle field is obtained through the method that speckle light intensity distribution on an original detection face is magnified and imaged onto a detection face, and the signal to noise ratio of the measurement matrixes is increased.

The invention discloses a non-stable signal multi-fractal feature extraction method based on dual-tree complex wavelet transformation. The steps include: performing integration processing on a non-stable signal to be analyzed; performing dual-tree complex wavelet transformation on the integrated signal, and using wavelet decomposition scale coefficients and detail coefficients to obtain fluctuation components of the signal under each scale; using the obtained wavelet coefficient of each scale to estimate the instantaneous frequency of each scale, and obtaining a time scale value of each scale; based on the scale values, performing segmentation on the fluctuation components under each scale; calculating a fluctuation function of each order of the signal, utilizing a double-logarithm relation of the fluctuation functions and the scale values, obtaining a generalized hurst index through least squares fitting, and obtaining scale index of each order; and utilizing legendre transformation to obtain a multi-fractal singular spectrum of the signal. The non-stable signal multi-fractal feature extraction method provided by the invention utilizes dual-tree complex wavelet transformation to perform signal decomposition, overcomes the problem that traditional wavelet transformation lacks translation invariance, ensures accuracy of multi-fractal feature extraction, the arithmetic speed is fast, and thus the method is in favor of online application.

The invention discloses a rectal cancer pathology image classification method based on a multi-channel collaborative capsule network. A dynamic routing mechanism of a capsule network is utilized to construct a network model, a multi-input feature fusion mode is utilized in a feature extraction layer to carry out feature extraction on a picture, a plurality of channels are arranged in parallel in the capsule layer to accelerate training, and then a margin loss function is utilized to train the model. The problems of spatial information loss, no rotation, translation invariance and the like in the training process of the traditional convolutional neural network technology are effectively overcome; and the generalization ability of the model is improved by utilizing feature fusion, and the capsule layer allows cooperative cooperation between channels to accelerate training so that the time cost can be effectively saved, the parameters of the network are reduced and the network training ismore efficient.

The present invention relates to an image retrieval method. According to the method, color features of images are extracted by using a color histogram, two features, such as a color vector of the color histogram and the height of a color column, are used as retrieving bases, the degree of similarity is calculated by using a fuzzy membership function in a fuzzy set theory, the similarity is judged by using an alpha-level fuzzy relationship, meanwhile, texture features of the images are extracted by introducing non-subsampled contourlet transform (NSCT), the images are decomposed by using the NSCT, mean values and standard variances of subband coefficients in different levels and multiple directions are extracted as feature vectors which serve as indexes of images in an image library, the degree of similarity among the images is calculated by using the fuzzy membership function in the fuzzy set theory, powerful direction information is reserved after the images are decomposed due to the multi-scalability, multi-directionality and translation invariance of the images, thus, the texture features of the images can be described more comprehensively, and finally, the images are retrieved through combining two algorithms and applying comprehensive features.

The invention discloses a non-local means filtering method for a speckle noise pollution image. The method comprises that iterative computation of a neurons firing state image series of the speckle noise pollution image is conducted through a pulse transmission cortex model, a Renyi entropy vector is extracted by the neurons firing state image series, and non-local means filtering is conducted on the speckle noise pollution image based on the Renyi entropy vector, so that a denoised gray value is obtained. By the aid of the method, rotation invariance, translation invariance and scaling invariance can be extracted from the image containing speckle noise, the method can use more image information for denoising than traditional methods, besides, the similarity between two image pixel blocks can be calculated reasonably, image noise can be suppressed significantly, and the peak signal-to-noise ratio of the image can be improved, so that detailed information of the image can be effectively protected.

The invention discloses a drug detector standardization method based on the dual-tree complex wavelet algorithm. Firstly, multi-scale spectrum decomposition and reconstruction are conducted with the dual-tree complex wavelet, then each layer of reconstructed spectrum is corrected through piecewise direct standardization, predication models based on partial least squares and leave one cross validation are established, and finally all the predication models are fused and evaluated according to a calculated weight. According to the dual-tree complex wavelet and direct standardization combined method, due to the translation invariance and multi-scale property of the dual-tree complex wavelet, the defects of existing model transfer methods are overcome, the capacity of correcting drifting in the X-axis direction and Y-axis direction is excellent, and the method also has the advantages of being exquisite, precise and efficient and can be widely applied to the fields including near-infrared and Raman spectrum.

The invention relates to a point cloud description method based on projection contour signature and a distribution matrix. Firstly a local reference coordinate system having rotation and translation invariance is constructed by using the spatial coordinate information of the target point cloud, and then the target point cloud is transferred to the reference coordinate system. A space boundary boxof the target point cloud after attitude normalization is constructed, and the target point cloud is respectively projected to the three adjacent planes of the boundary box. The projection contour signature and the distribution matrix of projection of the point cloud on each projection plane are calculated for enhancing the expression ability of the descriptor for the visible part of the target point cloud, and the related series rules are determined according to the statistical features of point cloud projection. The sub-features of the target point cloud on the three projection planes are connected in series in the form of a histogram so as to obtain the final CSDM descriptor.

The invention provides an underwater sonar image denoising method based on dual-tree complex wavelet transform and PCA. The method comprises the following steps: performing the dual-tree complex wavelet transform to an underwater sonar image; converting the image from a space domain to a complex wavelet domain; maintaining a low-frequency approximate component obtained after the image is performed with three-layer dual-tree complex wavelet transform to be the same; processing a high frequency component of the image; using a PCA method to estimate noise energy in a high frequency sub-band so as to determine a threshold and using a hard threshold function to process a complex wavelet coefficient; performing the dual-tree complex wavelet transform to the processed complex wavelet coefficient so as to obtain the final denoising image. A traditional two dimension wavelet lacks translation invariance and direction selectivity. The method can overcome the above disadvantages. Image directivity information can be captured well. During the denoising, detail information, such as an image edge, texture and the like, can be protected.