140results about How to "Improve refactoring quality" patented technology

The invention discloses a non-convex compressed sensing image reconstruction method based on a redundant dictionary and structure sparsity. A reconstruction process of the method includes: observing original image blocks; using a mutual neighboring technology for clustering observation vectors; using a genetic algorithm for finding optimal atom combinations in a dictionary direction for each class of observation vectors, and preserving species; after species expansion operation is executed on each image block, using a clonal selection algorithm for finding an optimal atom combination on scale and displacement in a determined direction for each image block; reconstructing each image block by the optimal atom combination; and piecing all the constructed image blocks in sequence to form an entire constructed image. Image structure sparsity prior and redundant dictionary direction features are fully utilized, the genetic algorithm is combined with the clonal selection algorithm, and the method is used as a nonlinear optimization reconstruction method to realize image reconstruction. The reconstructed image is good in visual effect, high in peak signal noise ratio and structural similarity, and the method can be used for non-convex compressed sensing reconstruction of image signals.

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 self-adapting image compressive sampling method based on a multi-dimension saliency map, and is used for solving the problem of sampling resource waste due to the average allocation of the sampling rate to the image during compressive sampling. The method mainly comprises the steps of: carrying out support value transformation (SVT) on an sampled image and calculating toobtain a saliency map of the image; determining a vision salient region and a vision non-salient region according to the saliency map; allocating measurement data, allocating more sampling resources to the vision salient region; and reconstructing the measurement data obtained by self-adapting sampling through a nonlinear reconstructing algorithm to finally obtain an reconstructed image. Comparedwith the prior art, the method has the advantages that: when the compressive measurement of the image is carried out, according to the difference in people vision attention regions, self-adapting sampling resource allocation can be achieved based on different attention regions, thus the utilization rate of the sampling resources is increased and the quality of the recovered image is improved simultaneously. The method can be used for self-adapting compressive sampling of natural images, remote sensing images and the like, and has broad application prospects in low-cost imaging equipment.

The invention discloses a three-dimensional model reconstruction method and device, equipment and a storage medium, and relates to the field of three-dimensional modeling. The method comprises the steps of obtaining a single target image, wherein the target image comprises an image of a target reconstruction object; performing feature extraction on the target image to obtain a target feature map corresponding to the target image; inputting the target feature map into a grid reconstruction network; the target feature map is input into a voxel reconstruction network, the grid reconstruction network and the voxel reconstruction network are in semantic connection, the grid reconstruction network is used for reconstructing a three-dimensional grid model of a target reconstruction object according to the target feature map, and the voxel reconstruction network is used for reconstructing a voxel model of the target reconstruction object according to the target feature map; and constructing atarget three-dimensional model according to the target grid information output by the grid reconstruction network. In the embodiment of the invention, the three-dimensional model reconstruction of thetarget reconstruction object can be realized by only needing a single image, the requirement on a two-dimensional image required by three-dimensional reconstruction is reduced, and the three-dimensional reconstruction process is simplified.

The invention discloses a compressed sensing based color imaging device and a compressed sensing based color imaging method. The compressed sensing based color imaging device comprises a PC (personal computer), a DLP (digital light projector), a target image, a single-pixel photon detector and a data collection and control module. The PC, the DLP and the single-pixel photon detector are connected with the data collection and control module. The PC generates a two-dimensional random binary projected intensity image. The data collection and control module controls the DLP to perform binary intensity image projection on the target image under red, green, blue and white structured light sequentially to acquire digital signals of the red, green, blue and white structured light respectively and send the digital signals to the PC, reconstructs the signals of the red, green, blue and white structured light through compressed sensing to acquire four monochrome gray level images, and performing color fusion on the four monochrome gray level images to acquire a final color image. The compressed sensing based color imaging device and the compressed sensing based color imaging method have the advantages that light filters are not needed, and simple structure and low cost are achieved; sampling frequency can be reduced, and color fidelity can be guaranteed without reduction of image resolution.

The invention relates to a distributed compressed sensing video encoding and decoding method based on foreground-background separation, comprising the following steps: dividing a video according to the frame sequence into key frames (I frames) and non-key frames (P frames) based on the requirements of reconstruction accuracy and real-time performance; for the I frames, selecting an appropriate observation matrix to measure the I frames; for the P frames, performing foreground-background separation on the P frames and the corresponding I frames and then using the same observation matrix to measure the separated foreground, completing foreground-background separation in a differential manner, introducing decoding mode judgment at an encoder, classifying the decoding modes of the P frame in a GOP into three categories: a skip mode, a normal mode and a dictionary learning mode, and using the same observation matrix to measure the difference. By using the method, the complexity of the encoder can be reduced.

The invention discloses a depth image coding method which is characterized in that a maximum tolerable distortion distribution image of an original left viewpoint depth image is obtained according to influence of depth distribution on virtual viewpoint image drawing and by combining the visual characteristics of human eyes, a drawing image of an original right viewpoint depth image is obtained in the method based on depth image drawing, the original left viewpoint depth image is coded according to the maximum tolerable distortion distribution image and the set coding forecast structure to obtain a decoded left viewpoint depth image, the original right viewpoint depth image and a residual image of the drawing image of the original right viewpoint depth image are coded according to the set coding forecast structure to obtain a decoded residual image, and finally a reconstructed image of a decoded right viewpoint depth image is obtained according to a drawing image of the decoded right viewpoint depth image and the decoded residual image. The depth image coding method has the advantage of greatly improving depth image coding efficiency on the basis of ensuring virtual viewpoint image drawing performance.

The invention provides an image compression encryption method and a cloud auxiliary decryption method based on compressed sensing and optical transformation. The encryption method comprises the following steps: generating a chaotic sequence by respectively utilizing an LTS system and a 2D-LASM system according to a hash key and an initial key generated by a plaintext image P; dividing the plaintext image P into an approximate component LL and three detail components HL, LH and HH by using discrete wavelet transform; performing lossless encryption on the approximate component LL by using a chaotic sequence X1, and performing lossy encryption on the detail components HL, LH and HH by using chaotic sequences U1 and K1; combining the encrypted matrixes LL1 and H1 into a complex domain matrix LLH1, and performing double random phase encoding on the complex domain matrix LLH1 by using two random phase mask matrixes to obtain matrixes LL2 and H2; quantifying the matrix LL2 and the matrix H2;carrying out the random pixel scrambling on the quantized matrix by using a chaotic sequence Z1; recombining the scrambled matrixes to generate a sequence B, and diffusing the sequence B to obtain a ciphertext image C.

The invention provides a compressive sampling method, which includes: acquiring signals of a target object, selecting a sparse base capable of indicating the signals of the target object sparsely; generating quick projection operators according to the initial sparse base, a preset random permutation matrix and a random matrix, solving the optimum value of subcomponents according to the quick projection operators, and obtaining optimized sparse components; determining the sparse coefficient of the optimized sparse components in the initial sparse base, and reconfiguring to-be-sampled signals according to the sparse coefficient. Data in different data quantity can be acquired on blocks different in sparse degree by the quick compressive sampling method, and the random projection operators have lower computing complexity so that higher reconfiguring quality can be obtained. Besides, the invention discloses a compressive sampling device.

The invention discloses a compressed sensing image reconstruction method based on relevance vector grouping, which mainly solves the problems of inaccuracy and low robustness of compressed sensing image reconstruction. The realization process is as follows: 1) receiving an observation matrix and an observation vector; 2) obtaining an initial relevance vector by the observation vector and a sending matrix; 3) dividing the relevance vector into sub-relevance vectors according to the spatial neighbourhood relationship of wavelet coefficients; 4) adding a component in each sub-relevance vector and sequencing the components; 5) updating the reconstructed wavelet high-frequency coefficients and observation vectors on the basis of a Bayesian framework according to the sequencing order; 6) carrying out invert wavelet transform on the reserved low-frequency wavelet decomposition coefficients and the reconstructed high-frequency wavelet coefficients to obtain a reconstructed image. Compared with OMP and BEPA methods, the compressed sensing image reconstruction method based on relevance vector grouping disclosed by the invention has the advantages of high quality and good robustness of the reconstructed image, and can be used for reconstruction for natural images and medical images.

The invention discloses a time-space video compressed sensing method based on a convolutional network, and mainly aims at solving the problems that in the prior art, the video compression time-space balance and the video reconstruction real-time performance are poor. The scheme of the method comprises the steps that a training data set is prepared; a network structure of a time-space video compressed sensing method is designed; training and testing files are written according to the designed network structure; a network of the time-space video compressed sensing method is trained; and the network of the time-space video compressed sensing method is tested. The network of the time-space video compressed sensing method adopts an observation technology of simultaneous time-space compression are conducted simultaneously and a reconstruction technology of using 'time-space blocks' to enhance the time-space correlation, not only can real-time video reconstruction be achieved, but also the reconstruction result has the high time-space balance, the reconstruction quality is high and stable, and the network can be used for compressed transmission of a video and follow-up video reconstruction.

The invention provides a spectrum camera based on all-pass single-template complementary sampling and an imaging method for solving a technical problem that a conventional spectral imaging system is complex in structure and poor in spectral image reconstruction quality. The spectrum camera comprises a lens group, an encoding module, an observation module, and an image reconstruction processing module. The lens group is composed of a group of lenses. The encoding module uses an area array structure formed by random combination of multiple transmission array elements and reflection array elements, and forms an acute angle with the transmission direction of the spectral image, and is used for randomly encoding the transmission information and the reflection information of the spectral image. The observation module comprises a transmission observation module and a reflection observation module and is used for performing transmission observation and reflection observation on the encoded spectral image. The image reconstruction processing module simultaneously fuses two observation results to reconstruct the spectral image. The spectrum camera is simple in system structure, high in luminous flux, and short in exposure time, and can be used in the fields of mineral exploration, natural disaster monitoring, and military reconnaissance.

The invention discloses a method for image processing based on a matrix variable variation self-encoder so that a problem of damaging the spatial structure by image vectorization processing can be solved and thus image reconstruction, denoising and completion can be realized. Different from the traditional VAE, the method has the following advantages: the input, hidden layer features, latent variable distribution parameters of the model are described by using an inherent expression form -2D matrix of the image; with the definition of matrix Gaussian distribution and related properties, an explicit expression of an objective function of a novel model is deduced and then model parameters are calculated by using a stochastic gradient descent algorithm. In the model, modeling of the spatial structure and statistical information of the image data is carried out well because the involved modeling process is oriented at the matrix variable, so that the image reconstruction quality is enhanced, noises are removed, and the image completion is realized.

The invention provides an adaptive partition compression and perception-based video compression method. The method comprises two steps, namely the step of adaptively partitioning video images, and classifying and assigning sampling rates for various image blocks. During the step of adaptively partitioning video images, a gray difference value between the adjacent pixels of a reference frame image is adopted as a basis for block size segmentation, and a partitioning threshold value T is set. The gray average difference value between the adjacent pixels of a current region block is compared with the threshold value, and the adaptive blocks of the video image are partitioned based on the quad-tree algorithm. In this way, flat regions are effectively separated from detail regions and edge regions. On the basis of the adaptive partitioning operation, an inter-frame difference value for the DCT coefficients of video pixels is adopted as a basis for partitioning, and various image blocks diversified in size are divided into three types, namely quickly changing blocks, transition blocks and slowly varying blocks. Meanwhile, appropriate sampling rates are assigned to different types of image blocks. The method is good in video reconstruction quality and short in reconstruction time. Under the same condition, the video reconstruction quality and the reconstruction time of the above method are better than those of the video uniform partitioning, compression and perception processing method.

The invention relates to a streaming media video compression method based on inter-frame correlation. The method comprises: step one, video compressive sensing coding based on bidirectional predictionis carried out; to be specific, front-rear inter-frame correlation and intra-frame correlation are introduced in a video coding end and are used as sampling rates of non-reference-frame block image reasonable allocation; measurement vectors of corresponding image blocks of the reference frame and the non-reference frame are obtained by using a same projection matrix and the image blocks are classified into three classes based energy difference values: a basic invariant block class, a slowly changing block class and a fast changing block class; a preliminary classification result is correctedby using the intra-frame correlation; and then a final classification result is obtained, different sampling rates are allocated to different classes of image blocks to carry out sampling, and the measurement vector point number of the current image block is adaptively adjusted to obtain a measurement vector. Therefore, the classification accuracy is improved.

The invention discloses a three-dimensional wavelet sub-band sorting and code stream packet sealing method in the telescopic video code, wherein the wavelet sub-band in the time domain is sorted according to the requirement of the sequential decode, and the wavelet sub-band in the time domain is divided into different levels; the wavelet sub-band in the time domain of the same level is sorted according to the size of the transmission distortion MES value; then the code stream after being sorted is packed, then transmitted to a receiving end, when the receiving end does not reach the time limit of the decode, then performs the retransmission when finding the package missing, with the retransmitting time smaller than or equal to the largest retransmitting time; when the receiving end reaches the time limit of the decode, the package missing is not retransmitted. The method provides an effective code rate transmission control method, thereby providing correct code stream distribution for the video code and the transmission, providing the code stream organizing and transmitting way of high performance three-dimensional retractability self-adapting to the isomery, the network bandwidth wave property and the time delay change and the terminal diversity of the user receiving.

The invention discloses a nonconvex compressed sensing image reconstruction method based on local similarity and local selection. The method comprises the following steps: 1) carrying out observation and reception after an image is partitioned; 2)utilizing a local growth method to carry out clustering on observation vectors of all the image blocks; 3) carrying out population initialization on the image block corresponding to each kind of observation vector according to the scheme that the polyatom direction and monatom direction coexist; 4) utilizing an improved genetic algorithm to carry out crossing, variation and selection operation based on a local selection mechanism on the populations obtained in the step 3), reconstituting corresponding image blocks and obtaining optimal atom combinations; 5) utilizing a clone selection optimization algorithm to study the optimal atom combinations on the aspects of dimension and displacement; and 6) piecing the image blocks obtained in the step 5) together in sequence to obtain a complete reconstructed image, and outputting the complete reconstructed image. The reconstructed image is good in visual effect and high in peak signal to noise ratio, and can be used for nonconvex compressed sensing reconstruction of image signals under the condition of low sampling rate.

The invention discloses a coding aperture spectral imaging method based on self-adaptive dictionary learning, and solves the problem that the reconstruction quality of a spectral image is poor due tothe fact that a redundant dictionary constructed by a traditional method cannot effectively and sparsely represent the target image in the previous coding aperture spectral imaging process. The self-adaptive learning is performed according to the measured value to obtain the redundant dictionary, and the method is used for improving the quality of the reconstructed spectral image. The method comprises the following steps: firstly, transforming an original coded aperture spectral imaging framework, and adopting an overlapping block measurement mode; estimating a plurality of spectral image blocks by using a least square method, constructing a training sample set, and performing adaptive training learning by using the sample set to obtain a new redundant dictionary; substituting the new dictionary into an imaging framework to reconstruct a target spectral image; and finally, carrying out loop iteration on the process until an optimal solution is obtained. The constructed redundant dictionary can adapt to a target image, and the spectral image reconstruction quality in coded aperture spectral imaging is greatly improved.

The invention provides a remote sensing image super resolution method based on dictionary learning, comprising the following steps: S1, performing dictionary learning of corresponding feature types based on an image library of different feature types; S2, identifying the feature type of an original image; and S3, carrying out an image super resolution reconstruction process using a corresponding feature type dictionary. The invention puts forward a super resolution method based on dictionary learning using the idea of feature classification. As the number of atoms in a dictionary used in the reconstruction process is reduced greatly, the dictionary scale is reduced greatly. The proportion of effective atoms in the classification dictionary used is increased significantly, so that the number of effective atoms in the dictionary used in the reconstruction process is almost the same as a general dictionary, and the speed of reconstruction is improved significantly while the quality of reconstruction is ensured. The process only involves the adjustment of parameters, the computation efficiency is high, and the reconstruction quality of remote sensing images is high.

The invention discloses an adaptive rendering method using linear prediction, comprising the following steps: S1, defining prediction windows, and constructing multiple linear models through a recursive least-squares iteration process according to different sizes of the prediction windows; S2, calculating error caused by linear model prediction through recursive error analysis, and choosing the best prediction size for the prediction windows; and S3, distributing more light samples by use of an adaptive sampling method and according to the calculated prediction error to complete rendering. According to the adaptive rendering method using linear prediction provided by the invention, multiple pixels are reconstructed by independent linear models at the same time, costly error calculation is needed only at a small number of pixels, and therefore, the computational overhead is reduced greatly.

The invention discloses a coastal zone hyperspectral image distributed lossy coding and decoding method based on a region of interest, which belongs to the field of image processing. According to spectral correlation of the hyperspectral image, a spectral frame is divided into a plurality of spectral groups, and each spectral group comprises one key frame and a plurality of WZ frames; after region of interest extraction is carried out on the key frame of each spectral group, JPEG-LS lossless coding and decoding are carried out on the region of interest of the key frame, SPIHT-based coding and decoding are carried out on a background region, and compressed sensing-based coding and decoding are adopted for the WZ frames of each spectral group. The method of the invention has the advantages that the coding complexity is low, the error resilience is strong, the quality of a bit stream can be graded and the like.