44results about How to "Recovery Details" patented technology

The invention discloses a foggy day image sharpening method transmitted based on the self-adapting cluster color, including the following steps: collecting the source image and the target image information, to respectively perform the decoupling process through the color space; counting the mean and the variance of the source image and the target image to make the statistic characteristics of the source image 'close to' the target image as much as possibly; reverting the calibration source image obtained in the L alpha beta color space into the RGB color space through the L alpha beta color space, to obtain an image transmitting calibration result image; go on clustering the result image obtained through the color transmitting calibration and the target image; performing the similar finding and corresponding; performing the second color transmitting calibration; manually adjusting the number of cluster by second calibration to obtain the final calibration result image. The method of the invention realizes the sharpening process on the shot image in the fog, resumes the effective information in the source image.

The invention belongs to the technical field of computed tomography image processing. According to the specific technical scheme, the CT image super-resolution reconstruction method based on the generative adversarial network comprises the following specific steps: 1, establishing a dense connection relationship among different residual blocks based on a multi-stage dense residual block generatornetwork; 2, adding a bottleneck layer to the front end of each dense residual block; 3, optimizing the global network by adopting the Wasserstein distance loss and the VGG feature matching loss; 4, arranging a multi-path generator based on the sequence from thick to thin; 5, generating an image based on conditional expression generative adversarial learning; 6, reconstructing a CT image super-resolution reconstruction framework of the generative adversarial network based on multiple paths of conditions from coarse to fine; 7, reconstructing a loss function. According to the method, network redundancy is reduced, feature multiplexing among different residual blocks is realized, the maximum information transmission of the network is realized, the feature utilization rate is improved, and thereconstructed image quality is greatly improved.

The invention discloses a single image defogging method based on sub-pixel and conditional antagonism generation network, which comprises the following steps: obtaining an original fogless image dataset and synthesizing the fogged data set according to a foggy day imaging model; inputting the fogged image to be processed into a generator G, wherein the network structure of the generator G is provided with a skip layer connection, a feature map with gradually reduced encoding output size is encoded, and the feature map is respectively obtained by deconvolution and sub-pixel in a decoding stage, and then the feature map is operated by convolution to obtain a fogless image output by the generator; inputting the non-fog image and the original non-fog image output from the generator G into thediscriminator D, and judging whether the non-fog image output from the generator D is true or not; the generators G and the discriminator D are constrained by antagonism at the same time, and the antagonism loss and L1 loss are calculated. The parameters of the generators G and the discriminator D are updated by back propagation according to the principle of stochastic gradient descent. When thetotal loss of the model converges, the training of the model is completed.

The invention discloses an improved self-adaptive multi-dictionary learning image super-resolution reconstruction method, which comprises the steps of: (1) determining a downsampling matrix D and a fuzzy matrix B according to a quality degradation process of an image; (2) establishing a pyramid by utilizing the self-similarity of the image, regarding an upper-layer image and a natural image of the pyramid as samples of dictionary learning, constructing various types of dictionaries Phi k by adopting a PCA method, and regarding a top-layer image of the pyramid as an initial reconstructed image X<^>; (3) calculating a weight matrix A of nonlocal structural self-similarity of sparse coding; (4) setting an iteration termination error e, a maximum number iteration times Max_Iter, a constant eta controlling nonlocal regularization term contribution amount and a condition P for updating parameters; (5) updating current estimation of the image; (6) updating a sparse representation coefficient; (7) updating current estimation of the image; (8) updating a self-adaptive sparse domain of X if mod(k, P)=0, and using X<^><k+1> for updating the matrix A; (9) and repeating the steps from (5) to (8), and terminating iteration until the iteration meets a condition shown in the description or k>=Max_Iter.

The invention relates to a method for rebuilding a super-resolution human face image by position block nonlinear mapping. The method comprises the following steps of: segmenting the human face image for training and testing high resolution and low resolution into corresponding high-resolution position block images and low-resolution position block images which are provided with overlapped regionsfrom structural information and detail information of a roughly-aligned human face image; rebuilding super-resolution position block images corresponding to testing low-resolution position block images at each corresponding position by using a nonlinear mapping relation which is between the high-resolution position block images and the low-resolution position block images and learned from training data; and splicing the super-resolution position block images to obtain the final super-resolution human face image. According to experiments on a standard database, the method provided by the invention has a good visual effect and high objective estimation quality.

A method for image restoration and reconstruction registers each frame of an observed image sequence to suppress geometric deformation through B-spline based non-rigid registration, producing a registered sequence, then performs a temporal regression process on the registered sequence to produce a near-diffraction-limited image, and performs a single-image blind deconvolution of thenear-diffraction-limited image to deblur it, generating a final output image.

The invention discloses an image real-time super-resolution reconstruction method based on acceleration of a GPU. The image real-time super-resolution reconstruction method based on the acceleration of the GPU mainly solves the problem that an existing high quality image super-resolution reconstruction serial algorithm is hard to process in real time. The method comprises the following steps of (1) inputting a low-resolution image, (2) carrying out bicubic interpolation up-sampling on the low-resolution image to obtain an interpolation image, (3) carrying out image deconvolution operation based on consistent gradient and opposite direction constraint to the obtained interpolation image, and carrying out parallelization acceleration on the deconvolution operation on the GPU so as to obtain an output super-resolution image in ultrashort time. Experiment shows that the image real-time super-resolution reconstruction method based on the acceleration of the GPU can obtain high quality super-resolution results which contain good details and are clear in texture and natural and reasonable in structure in quite short computing time, and can be widely applied to relative application fields of changing of image resolution.

The invention relates to a video super-resolution recovery method based on deep learning and adjacent frames. The method comprises the following steps: (1) data preprocessing: (2) data enhancement: (3) data conversion: (4) network architecture construction: a network comprises a frame alignment module, a frame fusion module and a reconstruction module; (5) inputting each current video image frameneeding to be subjected to super-resolution recovery and n frames (2n + 1 frames in total) of images before and after the current video image frame into the network architecture constructed in the step (4); and outputting the recovered super-resolution video. According to the invention, the obtained image has a better effect.

The present invention discloses a heterogeneity-projection hard-decision interpolation method for color reproduction, which utilizes a heterogeneity-projection method to determine the optimal edge direction and then utilizes a hard-decision rule to determine the optimal interpolation direction and obtain the information of the green color elements. The high-frequency information of the plane of the green color elements is incorporated into the processing of the planes of the red color elements and the blue color elements to reduce the restoration errors of the red and blue color elements. Therefore, the present invention can decrease the interpolation-direction errors and achieve a higher PSNR and a better visual effect.

The invention discloses a sand and dust image clearing method according to an information loss restraint, wherein the sand and dust image clearing method comprises the steps of establishing an atmosphere light imaging model, a first step, calculating an atmosphere light component; a second step, calculating a rough estimated value of transmissivity and obtaining a rough transmissivity of a whole original image; a third step, thinning the rough transmissivity and obtaining the transmissivity of the whole image after thinning; and a fourth step, obtaining a finally recovered target image, and obtaining a target image after clearing based on an original sand and dust image. The sand and dust image clearing method provided by the invention realizes clearing processing on an image which is photographed in a sand and dust condition and realizes a relatively high clearing effect on sand and dust images with different degradation degrees and color shift degrees. The sand and dust image clearing method further has advantages of restoring effective information in the original image, enriching details, restraining halation, improving contrast and maintaining good scene color. The sand and dust image clearing method is suitable for image clearing in smog and sand-and-dust weathers with different degrees.

The invention discloses a real-time image defogging method based on an image simplified hierachical model, and the method is applied to cleanness processing of a real-time image under a foggy weather condition and belongs to the field of computer vision. The method comprises the following steps of: reading one frame video image from input foggy weather color video data; performing smoothing treatment on the video image, thus obtaining a basic image layer of the video image; using an original video image to reduce the basic image layer of the video image, thus obtaining a detail image layer ofthe video image; stretching the detail image layer of the video image, thus obtaining the enhanced detail image layer of the video image; adding the enhanced detail image layer of the video image with the basic image layer of the video image to obtain the defogged video image; and reading the next frame video image until all of the video image frames in the color video data are processed. The method can be used for effectively defogging and also can be used for reducing the complexity of the algorithm, thus the cleanness process of the real-time image under the foggy weather condition can be realized.

The invention relates to a prostate transrectal ultrasound (TRUS) image segmentation method based on a convolutional neural network. The method comprises the following steps: 1) extracting multi-scaleconvolution features in a prostate TRUS image by using an extended spatial pyramid pooling (DSPP) module with global feature coding semantic information; 2) in order to process the missing boundary in the segmented prostate shadow region, proposing a super-hybrid feature (SHF) capable of mixing low-level features and high-level features; and 3) a novel network composed of an encoder and a decoderis provided. The result shows that the method can achieve the precise segmentation of the prostate TRUS image, improves the segmentation accuracy and speed compared with the prior art, and successfully solves problems that the manual segmentation efficiency of the prostate TRUS image is low, the accuracy is low, and excessive manpower resources are consumed.

The invention discloses an underwater image restoration model based on multi-branch gating fusion and a restoration method thereof. The underwater image restoration model comprises a multi-branch feature extraction module, a gating fusion module and a reconstruction module which are connected in sequence. The multi-branch feature extraction module is used for extracting image feature information of different scales and different levels of an underwater original image to be restored; the gating fusion module is used for fusing the image feature information of different scales and different levels of the underwater original image to be restored to obtain an underwater low-resolution feature image; and the reconstruction module is used for carrying out image reconstruction on the underwater low-resolution feature image to obtain a high-resolution feature image, namely an underwater restored image. The multi-branch feature extraction module is composed of a plurality of different branch modules used for extracting image feature information of different scales and different levels of the to-be-restored underwater original image, the sizes of feature maps output by the different branch modules are the same, the contrast ratio is effectively improved, and the chromatic aberration is eliminated.

The invention relates to a multi-exposure-image high-dynamic-range imaging method and system based on a generative adversarial network. The method comprises the following steps: firstly, preprocessinga low-exposure image, a normal-exposure image, a high-exposure image and a reference high-dynamic-range image serving as a label with original resolution to obtain grouped low-exposure, normal-exposure, high-exposure and high-dynamic-range image blocks for training; designing a generator network for multi-exposure image high dynamic range imaging and a discriminator network for adversarial training; alternately training the generator network and the discriminator network to converge to Nash equilibrium by using grouped low-exposure, normal-exposure, high-exposure and high-dynamic-range imageblocks; and finally, inputting the low-exposure image, the normal-exposure image and the high-exposure image which are used for testing and have original resolutions into a converged generator networkto obtain a high-dynamic-range image prediction result output by the generator network. According to the invention, the quality of a high dynamic range image generated when obvious background movement or object movement exists among multiple exposure images can be improved.

The invention discloses a method and device for reconstructing a super-resolution facial image, and belongs to the field of image processing. The method comprises the step of dividing a tested facial image and a trained facial image into image blocks; the step of dividing the image blocks of the tested facial image into smooth blocks and non-smooth blocks; the step of continuing to divide each non-smooth block until there is no non-smooth block or the divided non-smooth blocks meet the preset conditions; the step of dividing the trained facial image into sub-blocks according to the same manner; the step of calculating reconstructed image blocks corresponding to all non-smooth sliding blocks in the tested facial image; the step of carrying out bicubic interpolation on all smooth blocks in the tested facial image to obtain corresponding reconstructed image blocks; the step of synthesizing the reconstructed images of the non-smooth blocks of the tested facial image and the reconstructed images of the smooth blocks of the tested facial image into a facial image to obtain the super-resolution reconstructed facial image of the tested facial image according to the position. The device comprises a dividing module, a self-adaptation module, a reconstructing module and a synthesizing module. According to the method and device, the definition of the reconstructed facial image is improved.

The invention relates to an image processing system and a processing method. The system comprises an image buffer, an image information analyzer, an image reconstruction device, a detail edge enhancement processor, a de-noising processor, a color processor and a synthesizer. The method comprises: extracting motion vector information, color region information and noise distribution information of images according to the sequence data of read original images; restructuring images according to the motion vector information of the original images and transmitting new image data together with the motion vector information, color region information and noise distribution information of the images into a color processor; using the detail edge enhancement processor; using the de-noising processor; and obtaining final processed data according to the image data subjected to edge enhancement, the de-noised image data and the data after color enhancement. The invention utilizes motion estimation and motion compensation to effectively restore image details, enhance colors and remove image noise.

The present invention discloses a heterogeneity-projection hard-decision interpolation method for color reproduction, which utilizes a heterogeneity-projection method to determine the optimal edge direction and then utilizes a hard-decision rule to determine the optimal interpolation direction and obtain the information of the green color elements. The high-frequency information of the plane of the green color elements is incorporated into the processing of the planes of the red color elements and the blue color elements to reduce the restoration errors of the red and blue color elements. Therefore, the present invention can decrease the interpolation-direction errors and achieve a higher PSNR and a better visual effect.

The invention belongs to the technical field of digital video processing, and particularly relates to a compressed video quality enhancement method based on an attention mechanism and time dependence.The method comprises the steps of constructing an FAM module based on an attention mechanism; constructing an inter-frame long-term time dependency relationship guided LDE feature fusion module; constructing a fine RSDE feature fusion module guided by an inter-frame short-term time dependence relationship; obtaining feature information endowed with different attention by using an FAM module according to contributions of a plurality of input continuous frames; extracting long-term time dependence information from the features between the adjacent frames by using an LDE module to obtain an intermediate result and feature information; and finally, selectively extracting a short-term time dependency relationship from the previous enhanced frame by combining the FAM module and the RSDE moduleto generate a final enhancement result. Experiments show that the visual quality can be enhanced on a test set containing various real scenes, and the objective quality evaluation index is greatly improved.

The invention belongs to the field of image and video processing, and in particular relates to an image and video up-sampling method based on gradient domain transformation, which is particularly suitable for processing cartoon images and videos. The method comprises the following steps of: firstly carrying out up-sampling and amplification on a low-resolution image, and then obtaining the gradient of a high-resolution image after gradient transformation to be used as a constraint reconstructed high-resolution image according to a defined gradient domain transformation method. The method provided by the invention can rapidly and effectively carry out up-sampling on the cartoon images or videos, the computation is simple, the image after reconstruction has natural color, fuzzy edges caused by interpolation become clear, details are restored to a certain degree, and the visual effect is better.

The invention discloses a multi-modal data fusion drivable area detection method based on point cloud up-sampling. The method mainly comprises two parts of space point cloud adaptive up-sampling and multi-modal data fusion drivable area detection, and comprises the steps of: registering the camera and the laser radar through a joint calibration algorithm, projecting the point cloud to an image plane to obtain a sparse point cloud graph, calculating edge strength information by using a pixel local window, and adaptively selecting a point cloud up-sampling scheme to obtain a dense point cloud graph; and carrying out feature extraction and cross fusion on the obtained dense point cloud picture and the RGB image to realize rapid detection of a drivable area. The detection method provided by the invention can realize rapid and accurate detection and segmentation of the drivable area.

The invention belongs to the technical field of image processing, and particularly relates to an HDR image generation method and device, a computer readable storage medium and terminal equipment. Themethod comprises the steps of firstly obtaining a to-be-processed single-frame original image, then processing the original image by using a preset deep learning network, and generating an HDR image corresponding to the original image; wherein the deep learning network adopts a network structure combining a lightweight Unet network and a residual error module, and the deep learning network is obtained by training based on a preset training sample set. The lightweight Unet network further reduces the number of layers of a lower sampling layer, the number of layers of an upper sampling layer andthe number of channels of a feature map, greatly simplifies the network structure, and better recovers detail information of a highlight area due to the fact that a residual error module can effectively extract high-frequency information in an original image, thereby improving the image quality. HDR image generation effect is greatly improved.

In order to solve the technical problems that a traditional lens-free camera image reconstruction method is relatively sensitive to noise and relatively low in system depth of field, the invention provides a lens-free camera image reconstruction method based on a coding mask and a Learned-TSVD algorithm. The method comprises the following steps: encoding a propagation process of light by using an encoding mask, converting an original large-scale system measurement matrix into a left system measurement matrix and a right system measurement matrix which are small in scale by utilizing the separable characteristic of the coding mask and a TSVD algorithm; thirdly, constructing neural network training to circularly train the left and right system measurement matrixes, and reducing an error of an approximate operation on a final result; and finally reconstructing an image through the TSVD algorithm and a regularization algorithm. According to the method, the learned system measurement matrixes are used for subsequent calculation, so that the noise influence resistance of the whole reconstruction process is higher; scene images at other distances can be well reconstructed by using the learned system measurement matrixes, and the problem of low depth of field of other reconstruction algorithms is solved.

The invention discloses a method for improving the resolution of an infrared image and application of the method, and belongs to the technical field of infrared image resolution improvement. The method is based on support of an improved block compressed sensing theory, and comprises the following steps: a, establishing an original compressed sensing image super-resolution basic model; b, introducing an image degradation model; c, establishing an improved compressed sensing image super-resolution model; d, establishing a super-resolution image reconstruction objective function based on an improved block compressed sensing theory; and e, optimizing the objective function, and obtaining a reconstructed high-resolution image signal. The method is applied to detection of power equipment faults. According to the method, image details can be recovered more effectively in a short time, the edge contour of the infrared image is enhanced, and image segmentation is facilitated. The application has the characteristics of convenience in fault area positioning, fault type identification and the like.

The invention provides an image defogging processing method and device, and a computer storage medium, and the method comprises the steps: carrying out the negation of each pixel value corresponding to each pixel point in a target channel of a to-be-processed foggy image, and obtaining a dark channel image of the foggy image, the method comprises the following steps: acquiring a foggy image, sequentially acquiring a global ambient light value and each transmissivity value corresponding to each pixel point according to a dark channel image of the foggy image, and acquiring a defogged image of the foggy image according to each transmissivity value corresponding to each pixel point, each initial pixel value corresponding to each pixel point in the foggy image and the global ambient light value. According to the invention, a good image defogging effect can be achieved.

The invention relates to a video compressed sensing reconstruction method based on an LSTM network and image group quality blind evaluation. The method comprises: a reconstruction end receiving a frame observation vector code stream; and combining to form continuous image group observation vectors, executing multi-frame joint iterative reconstruction based on the LSTM network on each image group observation vector to obtain a corresponding reconstructed image group, outputting final reconstructed frames one by one, and determining whether to update a parameter set of the LSTM network accordingto a continuous condition that the number of iterations reaches a maximum value. According to the method, sparse prior modeling and a data driving mechanism can be combined, and the quality of the reconstructed video can be improved.

The invention discloses a hyperspectral compressed sensing method, device and system based on a modified linear hybrid model. The method comprises the steps: S1, acquiring observation data Y = AX after spectral dimension compression sampling of an original hyperspectral image are acquired, wherein A is an observation matrix, and X is the original hyperspectral image to be reconstructed; s2, constructing a modified linear hybrid model: X = ES + BEH, wherein E is an end member matrix, S is an abundance matrix, B is a correction matrix of the end member E, H is an abundance matrix corresponding to the corrected end member; respectively estimating optimal values of B, S and H based on the observation data Y; and S3, substituting the estimated values of B, S and H into the modified linear hybrid model to reconstruct an original hyperspectral image. A correction item BEH is introduced into the correction linear hybrid model, self-adaptive correction can be carried out according to the disturbance condition of each point on the spectrum, the hyperspectral image characterization capability of the model is improved, and the reconstruction quality of hyperspectral compressed sensing is further improved.

The invention provides an underwater image enhancement method based on a multi-scale dense generative adversarial network and a training method of a network model, the multi-scale dense generative adversarial network model comprises a generator network and a discriminator network, and the generator network comprises a jump connection block composed of a multi-scale dense feature extraction MSDB module. In addition to unsaturated loss, L1 distance and gradient loss, the model also adds VGG perceptual loss to enhance the structure of the UWGAN, the true value of the ground can be more clearly reserved through the unsaturated loss, the L1 distance and the gradient loss, the perceptual loss is used for comparing the difference of image feature space, so that the network can better recover the detail information of the underwater image, and the accuracy of the image feature space is improved. And underwater image enhancement is carried out to increase the definition of the image.

The invention discloses a remote sensing image pan-sharpening method based on a progressive cross-scale attention network. The method comprises the steps that 1, panchromatic images and multispectral images with different resolutions serve as input of sub-networks in all stages to be fused step by step; 2, constructing an intermediate reasoning layer for the pyramid in each stage, wherein the intermediate reasoning layer comprises a feature extraction stage, a feature fusion stage based on a cross-scale attention module and an image reconstruction stage; and 3, obtaining an optimal remote sensing image fusion sharpening network through training and optimization of the progressive cross-scale attention network, wherein the optimal remote sensing image fusion sharpening network is used for fusing any low-resolution multispectral image and panchromatic image to obtain a high-resolution multispectral image. According to the method, the correlation among the features on a plurality of specific scales can be captured, and the reconstructed image with a better visual effect is obtained progressively, so that the cross-scale feature related information is better mined, and the effect of the reconstructed image is improved.

The invention relates to a method for rebuilding a super-resolution human face image by position block nonlinear mapping. The method comprises the following steps of: segmenting the human face image for training and testing high resolution and low resolution into corresponding high-resolution position block images and low-resolution position block images which are provided with overlapped regions from structural information and detail information of a roughly-aligned human face image; rebuilding super-resolution position block images corresponding to testing low-resolution position block images at each corresponding position by using a nonlinear mapping relation which is between the high-resolution position block images and the low-resolution position block images and learned from training data; and splicing the super-resolution position block images to obtain the final super-resolution human face image. According to experiments on a standard database, the method provided by the invention has a good visual effect and high objective estimation quality.