111 results about "Super resolution algorithm" patented technology

The invention discloses a deep learning super-resolution reconstruction method based on residual sub-images; residual sub-images are effectively combined with deep learning method based on convolutional neural network, super-resolution reconstructed images are clearer, and reconstruction speed is higher. By increasing the depth of convolutional neural network, a network model acquired by learning has higher nonlinear expression capacity and image reconstructing capacity; in addition, by introducing residual sub-image process, preprocessing based on traditional interpolation algorithm is removed, and fuzzy effect due to the interpolation algorithm is avoided. By making ingenious use of residual sub-images, it is possible to transfer deep learning convolutional operation from high-resolution space to low-resolution space, and accordingly reconstruction efficiency of super-resolution algorithm is increased at the premise of improving super-resolution reconstruction effect.

The invention discloses a human face super-resolution algorithm based on regional depth convolution neural network. The algorithm comprises the following steps: a training stage: S1) dividing the mutually overlapping image blocks in the pixel domain of an inputted human face image with low resolution to obtain a plurality of local regions; S2) extracting the local regions for local characteristics; S3) performing non-linear change to the local characteristics to obtain non-linear characteristics; S4) processing the non-linear characteristics to obtain reconstructed image blocks with high resolution; S5) splicing the image blocks with high resolution; adjusting the multi-layer convolution layers and correcting the parameters of the linear unit layer; and a testing stage: S6) inputting the tested human face image with low resolution; processing through the super-resolution network to obtain the human face image with high resolution. The regional convolution neural network proposed by the invention improves the quality of subjective and objective reconstruction of reconstructing high resolution images.

A super resolution optofluidic microscope device comprises a body defining a fluid channel having a longitudinal axis and includes a surface layer proximal the fluid channel. The surface layer has a two-dimensional light detector array configured to receive light passing through the fluid channel and sample a sequence of subpixel shifted projection frames as an object moves through the fluid channel. The super resolution optofluidic microscope device further comprises a processor in electronic communication with the two-dimensional light detector array. The processor is configured to generate a high resolution image of the object using a super resolution algorithm, and based on the sequence of subpixel shifted projection frames and a motion vector of the object.

An image shifting apparatus may capture multiple frames of an image, with each frame offset by a sub-pixel offset distance. After capture, the various sup-pixel offset images may be registered together and analyzed using known resolution enhancement algorithms. Super resolution algorithms, for example, may take the various frames and perform edge identification and peak extraction routines to improve the resolution of high frequency data within an image. The image shifting apparatus may include an actuation controlled optical compensator within the imaging system that may be adjusted to create a sub-pixel offset image of an original reference image. The optical compensator may be an image-stabilizing element, for example, capable of forming sub-pixel shifts in an image plane.

The invention discloses a video super-resolution reconstruction method based on a deep residual network. According to the method, a corresponding high-resolution image is reconstructed from a set of continuous low-resolution video frame images of a video sequence so that the video display effect can be obviously enhanced. The innovativeness of the video super-resolution algorithm is mainly reflected in two aspects: firstly, the initial stage, the series convolutional layer computation stage and the residual block computation stage are directly performed from the low-resolution video images by using the deep residual network and then the high-resolution video image is reconstructed by using the deconvolution and convolution operation mode gradually so that conventional preprocessing of bicubic interpolation does not need to be performed on the low-resolution video images; and secondly, compared with the most classic single frame and video super-resolution reconstruction method based on deep learning, the high-resolution video image can be effectively reconstructed in different environments under the condition of using few training data, and the video image display effect can be greatly enhanced.

The invention relates to an image super-resolution method based on overcomplete dictionary learning and sparse representation. The method comprises the following steps of: extracting two overcomplete dictionaries (a low-resolution image block dictionary and a high-resolution image block dictionary) in a large-scale dataset and utilizing the two overcomplete dictionaries to realize super-resolution reconstruction of image sparse representation. Simultaneously, in order to further improve the super-resolution effect of color images, the invention also proposes UV chromaticity super-resolution reconstruction based on super-resolution luminance information. The image super-resolution method has wide application prospect in the fields of video monitoring, medical imaging, remote sensing image and the like.

The invention provides a super-resolution method based on a convolutional neural network, and is aimed at searching a single-image super-resolution method which is fast in speed and high in restoration quality through a deep convolutional neural network. The invention provides a method of super-resolution reconstruction algorithm fusion more suitable for a single image to solve the problems that the time loss is great, hardware demands needed for realization of algorithms are high and the like. For features of different super-resolution algorithms, conventional super-resolution algorithms are analyzed and selected, so the selected super-resolution algorithms achieve advantage complementation after being fused, and the disadvantages for realization of the conventional super-resolution algorithms are overcome.

The invention belongs to the technical fields of statistical pattern recognition and image processing, in particular to a super-resolution method based on an artificial neural network. In the invention, the artificial neural network is used for expressing the function mapping relation among low-resolution images and high-resolution images. The method of the invention comprises the following concrete steps: creating a training set; establishing a BP neural network for training; bonding high-resolution images which are obtained by training according to the corresponding relation; and then, obtaining super-resolution images. The invention overcomes the disadvantage of time consumption of the original super-resolution algorithm based on manifold learning, and obtains better effect.

The invention discloses a hyperspectral image super-resolution reconstruction algorithm based on matrix structure sparse non-negative decomposition. According to the reconstruction algorithm, a low-spatial-resolution hyperspectral image and a high-resolution color image are united to reconstruct a high-resolution hyperspectral image, and the problem that an existing algorithm can not accurately restore the high-resolution hyperspectral image. The method comprises the realizing steps that (1) the low-resolution hyperspectral image and the corresponding high-resolution color image are input; (2) local and non-local self-similarity of the hyperspectral images is utilized for constructing a spectrum reconstruction target function based on the matrix structure sparse non-negative decomposition; (3) an alternating direction multiplier method is adopted for alternative solving to obtain an optimized spectrum material coefficient and a spectrum material base; (4) a matrix of the optimized spectrum material coefficient and a matrix of the optimized spectrum material base are utilized to reconstruct the high-resolution hyperspectral image. According to the method, the restored hyperspectral image is clearer, the image edge is sharper, and the spatial resolution of the hyperspectral image can be effectively increased.

The invention discloses a super-resolution reconstruction algorithm based on improved neighborhood embedding and structure self-similarity, comprising the following steps: first, the neighborhood embedding method is improved by use of structure similarity, more accurate high-frequency initial estimation is obtained, and an initial estimation algorithm based on neighborhood embedding is realized; and then, the local self-similarity and multi-scale structure similarity of a low-resolution image are used to construct a reconstruction constraint for the purpose of reconstructing high resolution, and a sparse representation dictionary is established. Compared with the prior art, on the basis that the algorithm put forward by the invention solves the problem that the learning-based super-resolution reconstruction algorithm of predecessors needs a lot of training sets, the neighborhood embedding method is improved, the method is adopted to solve the problem of inaccurate high-frequency initial estimation in a super-resolution algorithm based on local self-similarity and multi-scale similarity, and the super-resolution reconstruction effect of images is enhanced; and the saw-tooth effect and the ringing effect are suppressed better, and a reconstructed high-resolution image is closer to the real image and is of better subjective and objective quality.

A scene based non-uniformity, correction method super-resolution for eliminating fixed pattern noise in a video having a plurality of input images is disclosed, comprising, the steps of warping, each of the plurality of images with respect to a reference image to obtain a warped set of images; performing one of averaging and deblurring on the warped set of images to obtain an initial estimate of a reference true scene flame; warping the initial estimate of the reference true scene frame with respect to each of the plurality of images to obtain a set of estimated true signal images; performing a least square fit algorithm to estimate a gain image and an offset image given the set of estimated true signal images; applying the estimated gain image and estimated offset image to the plurality of images to obtain a clean set of images; and applying a super-resolution algorithm to the clean set of images to obtain a higher resolution version of the reference true scene frame.

The invention discloses an angle estimation method of a bistatic MIMO (Multiple-Input Multiple-Output) radar high-speed and high-maneuvering target. The angle estimation method comprises the steps of receiving an echo signal of a high-speed and high-maneuvering target by a receiving array of a bistatic MIMO (Multiple-Input Multiple-Output) radar; performing conjugate multiplication on echo of the receiving array and sending signals in different distant units; performing Fourier transform on data after performing conjugate multiplication in a fast time domain and a slow time domain in sequence; estimating a target speed according to a peak value in the step 3; extracting target slow time frequency domain components of different separation channels in a fast time frequency domain along with a target Doppler frequency value; splicing target frequency domain data in different distance gates to form virtual array data crossing a plurality of distance gates; and estimating sending angles and receiving angles of the targets by using a super-resolution algorithm. Through the angle estimation method, the influence on separation of MIMO radar channels caused due to high-speed and high-maneuvering movement of the target can be avoided; an effective virtual array can be formed by crossing the plurality of distance gates; and the problem of target angle parameter estimation of the bistatic MIMO radar under the high-speed and high-maneuvering target can be solved.

The invention discloses a method for improving video resolution through a convolutional neural network. According to the method, a video super-resolution reconstruction model based on the convolutional neural network is constructed according to a video image sequence, and in the model construction process, super parameters, including convolution kernel size, the number of neural network layers, etc. of the convolutional neural network are set based on the characteristics of the video image sequence; and then a single-image super-resolution method is utilized to generate image training weightsused for initializing weight parameters of the video model, redundant information among video frames is fully utilized, multiple frame video images are used as input of the convolutional neural network model, and finally a high-resolution video is obtained through an incremental iteration method. Through the method, the video super-resolution model is high in training speed and high in predictionprecision; and a comparison test verification result indicates that compared with other super-resolution algorithms, the peak signal-to-noise ratio and structural similarity of the images reconstructed through the method have comprehensive optimal results.

The invention discloses a high-resolution image prediction method based on a loss function constructed considering image texture information. According to the method, first, a connection weight and offset of an SRCNN (convolutional neural network) are randomly initialized, and network parameters are set; after training data is preprocessed, a high-resolution image pair training set and a low-resolution image pair training set are obtained; next, low-resolution images are input into a network framework, and high-resolution images output by the network are obtained; then, the loss function considering image texture information is adopted to perform error calculation, if the number of iterations is not reached, weight correction is performed, and finally a trained network is obtained; and ata test stage, the low-resolution images are input into the trained network to obtain predicted high-resolution images. Through the constructed loss function, pixel loss can be measured, image textureinformation loss also can be measured, the defect of an SRCNN super-resolution algorithm is overcome, and further improvement to the performance of the SRCNN algorithm is effectively realized.

The invention provides a gradable video coding system based on multi-scale online dictionary learning. A multi-scale training set establishing module based on layered sparsity is used for obtaining layered sparsity structures, in different scales, of an image through wavelet transformation. By means of a Gaussian differential filter set, direction energy is extracted so that primitive areas in the image can be obtained, and a multi-scale training set is generated by cutting out image blocks of the primitive areas. By means of an online dictionary learning module, it is ensured that dictionary atoms are iterated and optimized under low complexity according to the stochastic gradient descent method so that a sub-dictionary base corresponding to the multi-scale training set can be generated. For low-frequency video frames, a cross-scale video frame reconstruction module learns lost high-frequency information on different levels through the constructed sub-dictionary base; the aim of grading video quality is achieved through different-grade wavelet inverse transformation reconstruction. By means of the gradable video coding system, complexity of a super-resolution algorithm based on learning is lowered, the reconstruction quality gain is obtained at different transmission rates compared with H.264, and high expandability is achieved.

The invention discloses a face super-resolution method and device based on multi-view texture learning. The method belongs to the field of human face image super-resolution, and comprises the following steps: firstly, down-sampling a high-resolution human face image pair to a target low-resolution human face image pair, carrying out blocking operation on the target low-resolution human face imagepair, separating out mutually overlapped image blocks, and extracting facial texture multi-scale features by using a residual pooling module network; then, the extracted face multi-scale features aresent to a texture attention module, texture information is fused and compensated by calculating an attention map, the most similar features are collected, and the SR performance is more effectively improved. Finally, a feature map of the target view image is updated by feature fusion to produce a high resolution result. The network provided by the invention is superior to other latest face image super-resolution algorithms, and a face image with higher quality can be generated.

The invention provides a scalable video encoding system based on hierarchical structure progressive dictionary learning. The system comprises a system framework based on a hierarchical structure, a progressive dictionary learning module and a scalable video frame reconstructing module. According to the system, due to a scalable B frame prediction structure, reconstructed frames are added into dictionary training as reference frames of a finer layer, and the complexity of a super-resolution algorithm based on learning is reduced through a random gradient descending method. Through the system, consistency of video frame movement can be effectively kept, and meanwhile space and quality are scalable based on the system frame of the hierarchical structure.

The invention provides an image super-resolution algorithm based on context-related multi-task depth learning, Three depth neural networks are designed for capturing the basic information, the main edge information and the small detail information of the image, and then the neural networks are trained in a multi-task learning framework for context-related connection and unification. Given the input low-resolution image, the trained neural network will output the basic image, the main edge image and the micro-detail image respectively, and the final high-resolution image will be fused from thebasic image and the micro-detail image. The algorithm can recover high resolution (HR) images only from static low resolution (LR) images. Moreover, the structure of the recovered HR image is well preserved, and the structure information in the ideal HR image can be recovered as much as possible.

The invention provides a high-precision visual measurement method, device and system based on a bionic algorithm. The method comprises the following steps: establishing a mapping relation between a pixel size and an actual spatial geometric size of a to-be-measured object; Obtaining low-resolution images of the plurality of to-be-measured objects; Carrying out super-resolution reconstruction through a super-resolution algorithm based on a residual network; For the reconstructed image, extracting edge points by using a Canny edge detection operator, extracting corner points by using Hilbert transform, and carrying out edge tracking by using the edge points and the corner points as heuristic information through a fruit fly algorithm; And finally, obtaining a single-pixel edge by utilizing arelated mechanism, and calculating the spatial geometric dimension of the to-be-measured object. The device comprises a mapping module, an image acquisition module, a reconstruction module, an edge coarse detection module, a fruit fly detection module and a calculation module. The system comprises an objective table, a CCD camera, a two-dimensional workbench and the like. According to the invention, the field of view of single imaging is effectively expanded, the measurement cost is reduced, and the detection efficiency is improved.

The invention discloses a sparse MIMO-OFDM channel estimation method based on a space-time correlation of a channel. Channel frequency domain responses of different transmit-receive antenna to a pilot frequency in a received OFDM symbol are estimated; the channel frequency domain responses of the different transmit-receive antenna to the estimated pilot frequency are arranged according to a certain rule into a matrix; the matrix is processed through the super-resolution algorithm so that the multipath time delay number of the channel can be acquired; the matrix is processed according to the acquired multipath time delay number, so that multipath time delays of the channel are acquired; multipath gain corresponding to a time delay is acquired according to the acquired multipath time delays and the acquired matrix of the channel; a channel frequency domain response to a subcarrier at data is acquired according to the acquired multipath time delays and the acquired gain. In this way, the problem that the number of required pilot frequencies increases along with the number of antenna in the channel estimation process in a current MIMO-OFDM system is solved; meanwhile, the space-time correlation of the channel is used for further improving channel estimation accuracy, remarkably reducing system pilot frequency spending, and improving spectrum efficiency.

The invention relates to a local feature transformation based face super-resolution reconstruction method. The method includes: performing nonnegative matrix decomposition for a low-resolution sample library matrix so as to obtain a local feature expression of a low-resolution image; transforming local features to a global feature space by the aid of a transformation relation between the local feature expression and a sample space reconstruction coefficient; as for the inputted low-resolution image, acquiring possessed features of the low-resolution image, then transforming to a sample space so as to obtain a global feature, and using a high-resolution sample library to substitute for a low-resolution sample library so as to obtain a high-resolution image; and using the high-resolution image obtained by reconstruction as an initial value, using a maximum posterior probability frame for iterative optimization of the inputted low-resolution image so that better image reconstruction quality is obtained. A global face super-resolution algorithm based on transformation of the image local features to the global feature is provided, detail representation capability of the global face algorithm is enhanced, and objective image quality of the reconstructed high-resolution image is improved.

The invention belongs to the field of computer vision, relates to an image super-resolution algorithm guided by a reference image, and aims to greatly improve the running speed and the visual result compared with an existing algorithm. The invention discloses a reference image guided super-resolution method based on dense matching and adaptive fusion. The method comprises the following steps: establishing a training data set; aligning the reference image with the low-resolution image; inputting the low-resolution image and the aligned reference image into a convolutional neural network for fusion; setting the learning rate of the network and the weight of the loss function of each part, training the convolutional neural network by using a deep neural network framework PyTorch until the loss converges, and generating a training model; and performing image super-resolution by utilizing the generated training model. The method is mainly applied to computer image processing occasions.

Known super-resolution algorithms are inefficient due to a high degree of redundant calculations, require search operations such as block matching for finding nearest neighbors, and achieve only small magnification factors. An improved method for upscaling an image comprises steps of upscaling the image by pixel interpolation to obtain a coarse upscaled image, and, for each pixel of the coarse upscaled image, determining a nonlinear regression function for a patch of pixels around a current pixel of the coarse upscaled image and enhancing the value of the current pixel by adding the result of the nonlinear regression function, wherein a pixel of an upscaled image is obtained. The nonlinear regression function is obtained from a trained regression tree, based on geometric features of the patch.

The invention provides a resolution compensating device and a resolution compensating method applied to three-dimensional (3D) image display and a 3D television. The device comprises an acquiring module for acquiring an observation low resolution (LR) image group according to a time frame sequence, a setting module for taking a first image in the observation LR image group as the current image to be compensated, and a resolution compensating module for compensating the resolution of the current image to be compensated by using a super-resolution algorithm. By using the resolution compensating device and the resolution compensating method applied to the 3D image display and the 3D television which are disclosed by the embodiment of the invention, the resolution of the acquired compensated images is further improved, so that the watching experience of users can be improved when the users watch television programs consisting of the compensated high-resolution images.

The invention discloses a real-time video super-resolution processing method integrated with cortex-A7. The method includes that (1), video sampling is performed, and a low-resolution video frame is obtained and input to a system on chip (SOC); (2) the low-resolution video frame is sequentially subjected to complexity processing, feature vector extracting and sample set training, feature vectors to be matched are obtained, and a sample set is established by high-resolution high-frequency components; (3) according to the feature vectors to be matched, the low-resolution video frame is subjected to super-resolution processing by means of an improved super-resolution algorithm based on cluster dictionary self-learning and feature sparse representation and combination with SOC coding and decoding technologies, and thereby, a high-resolution video frame flow is output. The real-time video super-resolution processing method integrated with cortex-A7 has the advantages of real time, low distortion rate and processing costs and high processing speed and qualities, and the method can be widely applied to the field of video image processing.

The invention provides a face super-resolution image processing method based on double-manifold alignment. Two heterogeneous manifolds of training-integrated high-resolution images and low-resolution images are subjected to double-manifold alignment in the space between a global face and a residual face, and then super-resolution algorithm is carried out. The invention has the advantage that the two heterogeneous manifolds of high-resolution images and low-resolution images are aligned by using Procrustes analysis, thereby improving the super-resolution effect of the images by the learning algorithm.

The invention discloses an image super-resolution reconstruction method based on a dense feature fusion network. The method comprises the following steps: 1) preprocessing data; 2) establishing an image super-resolution reconstruction model; And 3) inputting the to-be-processed image into the model to obtain a high-resolution image. The image super-resolution reconstruction model comprises a coarse feature extraction network, a dense feature fusion network and an image reconstruction network; the coarse feature extraction network is used for extracting coarse image features of a low-resolutioncolor image; the dense feature fusion network is used for extracting high-order image features from the coarse image features; And the image reconstruction network is used for adding and fusing the coarse image features and the high-order image features to obtain dense image features, and then reconstructing the dense image features to obtain a color high-resolution image. According to the method, noise caused by a traditional interpolation amplification super-resolution algorithm can be effectively reduced, more high-frequency information is obtained to realize high-resolution image detail restoration, and the precision of super-resolution reconstruction is improved.