99 results about "Perceptual image" patented technology

Provided are systems, methods and techniques which use local image properties to determine which potential defects in an image should be corrected. In one representative embodiment, potential defects in an image are identified based on edge sharpness, and measures of strength for the different potential defects are calculated. The measures of strength for the potential defects are then evaluated in view of certain local properties of the image, in an immediate vicinity of such potential defects, in order to identify the image defects to be corrected.

A system and method provide enhanced perceived contrast within a region of interest (102) of an image being displayed in order to enable viewers to more easily perceive subtle features in images, such as during medical diagnostics. The enhanced perceived contrast may be implemented using software only, hardware only, or a combination of hardware and software. If the backlight of a display includes only one or few light sources, the enhanced perceived contrast can be achieved through a software only solution in which the area (104) outside of the ROI (102) is darkened. If the backlight of the display comprises, for example, an LED matrix, enhanced perceived contrast may be implemented through a hardware only solution in which the luminance of the display is increased in the ROI (102). In addition, enhanced perceived contrast may also be achieved through a solution utilizing both hardware and software.

Three-dimensional imaging without parallax barriers or specialized eye gear, and without attendant loss of resolution, is provided by a display that produces dynamic images for display on at least two stacked electronic transmissive displays to create a continuous 3-D image field in a large viewing area or in multiple viewing areas. The images on each display are derived from stereoscopic image sources corresponding to both eyes of a viewer, and the derived images act as a mask for each other causing 3-D perception. The derived images are processed by summing the predicted image data, comparing the predicted image data to the desired stereopair, and minimizing the error. In preferred embodiments, the processing can be performed by an artificial neural network. A viewer may be presented with different aspects of an image as their viewing position changes to allow the viewer to perceive various perspectives of an image in dynamic fashion.

Methods and apparatus for providing optimized gamma settings for each of a plurality of viewing angles and / or device orientations. In certain types of display devices, off-axis viewing leads to contrast degradation and / or color aberrations in a perceived image, as luminance values depend on the angle at which the output is viewed. By remapping grayscale and / or color values to new output voltages, an image can be presented at an optimized luminance level when viewed from any specific angle. In some embodiments, the display device comprises an inclination sensor adapted to sense device rotation about at least one axis. Display parameter optimization logic reads data from the inclination sensor and automatically adjusts the display to an optimized gamma setting.

A perceptual coder is disclosed for encoding image signals, such as speech or music, with different spectral and temporal resolutions for redundancy reduction and irrelevancy reduction. The image signal is initially spectrally shaped using a prefilter. The prefilter output samples are thereafter quantized and coded to minimize the mean square error (MSE) across the spectrum. The disclosed perceptual image coder can use fixed quantizer step-sizes, since spectral shaping is performed by the pre-filter prior to quantization and coding. The disclosed pre-filter and post-filter support the appropriate frequency dependent temporal and spectral resolution for irrelevancy reduction. A filter structure based on a frequency-warping technique is used that allows filter design based on a non-linear frequency scale. The characteristics of the pre-filter may be adapted to the masked thresholds, using techniques known from speech coding, where linear-predictive coefficient (LPC) filter parameters are used to model the spectral envelope of the speech signal. Likewise, the filter coefficients may be efficiently transmitted to the decoder for use by the post-filter using well-established techniques from speech coding, such as an LSP (line spectral pairs) representation, temporal interpolation, or vector quantization.

The invention relates to a multi-resolution modeling method for three dimensional model of complex building, comprising the steps of: building corresponding semi-side data structure with meaning information in a memory at same time of opening the three dimensional model of the building; calculating viewpoint spherical surface when the building three dimensional model is maximized to screen space by testing minimum bounding ball of the dimensional model of the building; acquiring viewpoint sampling position, and rendering the dimensional model of the building to obtain initial rendering image;executing HVS filter based on two dimension wavelet decomposition to the initial rendering image, and acquiring sensed image through wavelet coefficient reconstruction; calculating difference image according to the initial rendering image and the sensed image, dividing the pixels with changed gray scale in the difference image to obtain sensing information; finding out triangles corresponding to pixels with the changed gray scale on surface of the dimensional model of the building, simultaneously transmitting the corresponding sensing information to the triangles; simplifying the model according to the sensing information of the triangles.

The invention discloses a compression perceptual image reconstruction method based on Generative Adversarial Networks. The method comprises the following steps: S1, according to a measurement vector obtained by original image sampling and a reconstruction image size, constructing a generation antagonism network model based on a neural network, and designing an objective function for optimizing thegeneration antagonism network model parameters; S2, presetting parameters when training the generated antagonistic network model; 3, alternately training a generator and a discriminator by adopting aback propagation algorithm accord to that objective function; 4, if that Generative Adversarial Network model converges, the train network can directly realize the compression sensing task, and the model output is the corresponding original image reconstructed by the measurement vector; Otherwise, return to Step S2-S4. The invention utilizes the powerful mapping ability of the generator to initially reconstruct the original image, and utilizes the confrontation training of the generator and the discriminator to make the pixel distribution of the image reconstructed by the generator closer tothe original image, thus achieving the purpose of accurately reconstructing the original image under the low sampling rate.

In one respect, provided are systems, methods and techniques in which local regions within an image are processed to provide fuzzy classification scores, which are calculated by determining changes in pixel values along a number of different directions. The resulting fuzzy classification scores are then used to detect or identify edge-containing or texture-containing regions, or to otherwise process the image regions differentially according to their fuzzy classification score. In another respect, provided are systems, methods and techniques for differential processing of different areas in an image. The differential processing in this case is based on calculated measures of local activity, which indicate features in corresponding local regions, and also based on calculated measures of local pixel-value variations, which indicate an amount of variation in pixel values across the corresponding local regions.

The invention discloses a multi-layer residual coefficient image coding method based on compressed sensing. The method comprises the steps of dividing noise into 17 layers and training the 17 layers; substituting the 17 newly trained denoising models into an LDAMP iteration algorithm to complete compressed sensing image reconstruction; carrying out compressed sensing reconstruction on each layer and predicting a lower-layer measurement value to obtain a residual error, and carrying out quantization by adopting a block adaptive quantizer with the same quantization bit depth in the coding method; wherein the information needing to be transmitted by the current layer of the image coding end is the difference value between the quantization result of the real measurement value and the quantization result of the prediction measurement values of the reconstructed images of the measurement values of all previous layers of the layer; and the current layer of the image decoding end receiving the transmission information of the corresponding layer of the coding end, obtaining a measurement value or a residual coefficient after adaptive arithmetic decoding, and the image reconstruction of the current layer of the decoding end utilizing the received and recovered image measurement values of all previous layers. According to the multi-layer residual coefficient image coding method based on compressed sensing provided by the invention, the rate distortion performance of image coding based on image compressed sensing reconstruction can be effectively improved.

Provided are systems, methods and techniques which use local image properties to determine which potential defects in an image should be corrected. In one representative embodiment, potential defects in an image are identified based on edge sharpness, and measures of strength for the different potential defects are calculated. The measures of strength for the potential defects are then evaluated in view of certain local properties of the image, in an immediate vicinity of such potential defects, in order to identify the image defects to be corrected.

A method for combining a random set of video features non-linearly to evaluate perceptual quality of video sequences includes (a) receiving a video sequence for image quality evaluation; (b) providing an objective metric image quality controller comprising a random set of metrics ranging from M1 to Mn without dependency information for each one metric; (c) applying each one metric individually to the video sequence to provide an individual objective scoring value of the video sequence ranging from x1 to xn; (d) determining a plurality of sets of weights (w1 to wn) which correlate to predetermined subjective evaluations of image quality for a predetermined plurality of video sequences (n), each one set of weights being assigned a range having an incremental value equal to the range divided by a number of combinations for each one set of weights; (e) weighting each individual objective scoring value x1 to xn provided by each one metric of the random set of metrics in step (c); (f) combining metrics of the weighted individual objective scoring value of the random set of metrics into a single objective evaluation F, wherein each weighted individual scoring value from step (e) is multiplied by each individual objective scoring value x1 to xn from step (c); (g) calculating a correlation factor R to provide a correlation value for the objective evaluation F and the plurality of video sequences (n). Steps (e), (f) and (g) are repeated to provide a plurality of correlation factors which are ranked. A heuristic search uses a genetic algorithm to find the best set of weights to provide objective scores closest to predetermined subjective evaluations. A system provides the hardware and modules that perform the non-linear combination of metrics to provide enhanced perceptual image information.

The invention discloses a multidimensional image-based product design scheme evaluation method. The method includes the following steps that: (1) a k-mean clustering analysis method and factor analysis method are utilized to obtain a typical sample and typical perceptual image which accord with the perceptual cognition of a user, and then, a user-image initial evaluation value is obtained through using a semantic differential method; and (2) a gray correlation analysis and technique for order preference by similarity to an ideal solution (TOPSIS)-combined method is used to realize order optimization of alternative design schemes under the specific requirements of the multidimensional image of the user, and this process includes determining an alternative design scheme set, generating an original evaluation matrix, calculating the weights of evaluation indexes through using information entropies, constructing a normalized evaluation matrix, calculating distances, calculating relative proximities and sequencing the schemes. With the method of the invention adopted, the subjectivity of the perceptual image acquisition of the user can be balanced, and the validity of data can be ensured.

The invention discloses a continuous action online learning control method and system for an automatic driving vehicle. The continuous action online learning control method comprises the following steps: encoding a perceptual image It through a deep encoding network to obtain an encoding state feature st; respectively inputting encoding state features st into actuators-actuators, wherein the evaluator models all adopt an evaluator network and an actuator network of a cerebellar model neural network, an action at is output through the actuator network, and an actuator is updated through the evaluator network; parameters of an evaluator model. According to the invention, a synthetic deep neural network feature coding technology and an enhanced learning principle are adopted; the learning control problem of a continuous action space is solved under high-dimensional state input; on-line learning control of a continuous action space under large-scale continuous state input can be realized,the learning period is shortened while the learning effect is ensured, the learning process can be quickly converged to obtain a control strategy with a good performance effect, and the data utilization rate is good.

The embodiment of the invention discloses a vehicle pose correction method and device, and the method comprises the steps of carrying out the particle sampling on the vehicle position information based on the prior position of a vehicle body when a coverage region corresponding to the prior position of the vehicle body is detected in a preset navigation map, wherein the prior position is obtainedthrough a preset positioning device; updating the poses of the particles obtained by sampling and the weight information corresponding to each particle so as to enable the positions of the target particles with the set number to meet a preset convergence condition; determining the state quantity of the vehicle body position according to the updated weight information of each target particle, and obtaining a target matching relationship between the perception image and the preset navigation map based on the state quantity and the vehicle body posture; and optimizing the pose of the vehicle bodyat the prior position based on the target matching relationship. By the adoption of the technical scheme, the problem that the positioning precision is not high when a consumer-level preset positioning device is used is solved, and the technical effect of the centimeter-level high-precision positioning of the vehicle is achieved.

A method for modifying images for display on a three dimensional display includes receiving a first image and a corresponding second image which together represent a three dimensional image when displayed on the display. Glasses suitable to be worn by a viewer of the display provide depth perception of the three dimensional image. The glasses selectively send a signal to the display indicating whether the viewer selectively observes either two dimensional or three dimensional content by the display. The glasses also selectively send a signal to the display indicating the depth strength and image brightness of the perceived images.

The invention discloses a multi-camera panoramic image construction method based on compressed sensing and super-resolution reconstruction. An image super-resolution theory is introduced into compressed sensing reconstruction; the method starts from the observation interference angles of reducing compressed sensing observation redundancy and removing damaged observation values. And a robust compressed sensing image super-resolution reconstruction technology is established, and a high-resolution and high-quality image with rich details from a damaged observation value in the set at a super-resolution far less than the observed quantity is reconstructed under the traditional compressed sensing theory under the constraint of the constructed minimum effective observation set. According to themethod, a novel compressed sensing image super-resolution reconstruction technology is applied to a multi-view camera network with limited power supply; under the condition that the resolution of an imaging system is low, reduction of the image acquisition data volume and improvement of the image spatial resolution can be achieved, the reconstructed panoramic image is rich in detail and high in resolution, and a theoretical basis is provided for further achieving long-time video monitoring of large scenes such as the field.

The embodiment of the invention discloses a method and device for updating a matching relationship between a navigation map and a perception image. The method comprises the steps of obtaining the current position of a vehicle provided by a preset positioning device, and correcting the elevation of the vehicle provided by the preset positioning device based on lane line information corresponding tothe current position in the navigation map; arranging and combining the traffic sign groups, meeting the set distance requirement with the current position, in the navigation map and the perception image collected by a camera so as to determine a target traffic sign matched with the position from the navigation map and the perception image; and correcting the pose of the vehicle in the navigationmap according to the position of the target traffic sign, and updating the matching relationship between the navigation map and the perception image based on the corrected elevation and the pose of the vehicle. By the adoption of the technical scheme, the problem that the positioning precision is not high when a consumption-level preset positioning device is used is solved, and the high-precisionpositioning effect also can be achieved through the consumption-level preset positioning device.

The invention relates to a compression perception image reconstruction algorithm based on depth learning. The method comprises the following steps: S1, preprocessing image data, including extracting gray value of the data and dividing the image into blocks; S2, measuring the segmented image blocks to obtain a measurement matrix; S3: Constructing a 10-layer deep compression perceptual reconstruction network; S4, training the 10-layer network in the depth learning framework; S5, after passing through that depth neural network, obtain the reconstructed image block, and rearranging the image blockaccording to the original row and column value accord to the index; S6, after that image blocks are rearrange to obtain a reconstructed image, a BM3D denoiser is selected to carry out denoising processing on the image, and finally the reconstructed image is obtained. The compression perception image reconstruction algorithm provided by the invention consumes most of time in the network training stage, and the image reconstruction speed is very fast after the network training is completed. The invention replaces the traditional reconstruction algorithm through the depth learning network, but still has good reconstruction accuracy.

The invention provides a display panel, a graphical structure is arranged in a film layer of the display panel, the display panel is at least provided with a first position and a second position different from the first position, and the first position and the second position meet preset conditions. As the film layers passing through the first position and the second position meet the corresponding relation, the phase difference is small after the light is emitted from the display panel through the two paths. Phase difference generated after light of the same phase passes through the display panel is one of the important reasons of diffraction. According to the invention, after the light with the same phase penetrates through the display panel through two paths, the phase difference is within a preset range, the diffraction phenomenon caused by the phase difference is reduced, the definition of an image sensed by a camera behind the display panel is improved, a clear and real image canbe obtained by a photosensitive element behind the display panel, and full-screen display is realized.

In one respect, provided are systems, methods and techniques in which local regions within an image are processed to provide fuzzy classification scores, which are calculated by determining changes in pixel values along a number of different directions. The resulting fuzzy classification scores are then used to detect or identify edge-containing or texture-containing regions, or to otherwise process the image regions differentially according to their fuzzy classification score. In another respect, provided are systems, methods and techniques for differential processing of different areas in an image. The differential processing in this case is based on calculated measures of local activity, which indicate features in corresponding local regions, and also based on calculated measures of local pixel-value variations, which indicate an amount of variation in pixel values across the corresponding local regions.

A content-aware image retargeting technique uses an “importance filtering” technique to preserve important information in the resizing of an image. The image saliency is first filtered, guided by the image itself to achieve a structure-consistent importance map. The pixel importance is then used as the key constraint in computing the gradient map of pixel shifts from the original resolution to the target resolution. Finally the shift gradient is integrated across the image by a weighted filtering process to construct a smooth pixel shift-map and render the target image. The weight is again controlled by the pixel importance. The two filtering processes enforce the maintaining of structural consistency while preserving the important contents in the target image. The simple nature of the present filter operations allow for real-time applications and easy extension to video retargeting, as the structural constraints from the original image naturally convey the temporal coherence between frames.

The invention relates to the technical field of image processing and provides a straight line-based different-source image coupling method. The method includes the steps of extracting line characteristics of a reference image and line characteristics of a sensing image, and taking the line characteristics as elements of different-source image coupling; constructing a neighborhood relation betweenline end points and end points and a neighborhood relation between end points and lines, connecting the end points with the lines with the neighborhood relations, obtaining straight lines with betterstability, and constructing line sets accurately representing scene structures; constructing line pairs between the line sets of the reference image and the line sets of the sensing image, and calculating possible transformation parameters; and selecting transformation parameters that enable the sensing image and the reference image to have the highest similarity, and thus realizing coupling of the sensing image and the reference image. A phenomenon of double sides can be effectively prevented when road lines are extracted from an aviation image, the number of lines for coupling is greatly reduced, and the geometrical structure of a scene can be represented by as few lines as possible. The method is easy to realize, is fast in calculating speed and can realize quantitative analysis.

The invention relates to a stereo image quality evaluation method based on a local multivariate Gaussian descriptor. The method comprises the following steps: establishing an absolute parallax image according to a stereoscopic parallax perception principle of a human and a parallax matching algorithm; simulating a binocular fusion and binocular competition mechanism of a human eye, filtering left and right views by use of a two-dimensional Gabor filter, and calculating weight factors of right and left viewpoints according to a Gabor response and binocular parallax; synthesizing a stereoscopic perception image; performing 5*5 partitioning on the stereoscopic perception image, performing DCT on each image block, and reorganizing images after the DCT; selecting center coordinates of each image block as coordinates of feature points, and calculating the contrast of each image block; calculating local multivariate Gaussian features of the image; and performing stereo image quality prediction.

The invention provides an edge perception image semantic segmentation method based on adaptive feature fusion. The edge perception image semantic segmentation method is a new semantic segmentation method based on a residual network. A double-branch network structure model comprises an edge branch and a semantic branch, the edge branch is led out from a shallow layer part of the semantic branch, and the semantic branch adopts a coding and decoding structure. In the edge branch, the added multi-scale cross fusion operation obtains image multi-scale features by superposing hole convolution with different hole rates; and meanwhile, the robustness of the multi-scale features can be further improved through cross fusion among the branches. In the semantic branch, the deep features and the shallow features are fused based on a spatial attention mechanism; a large amount of noise contained in the superficial layer feature can be filtered while rich spatial information contained in the superficial layer feature is obtained; finally, the semantic branch features and the edge branch features are fused, and the segmentation effect is further optimized.

The invention discloses a super-resolution graph recovery method for simultaneously enhancing underwater images, which is based on a generation model of a residual network, recovers a perception imagewith spatial resolution of 2 times, 3 times or 4 times through a learning algorithm, and supervises the training of the perception image by formulating a multi-mode target function. The objective function can solve the problems of color degradation, insufficient image definition and loss of advanced feature representation in a chrominance-specific underwater environment, and can also supervise and learn a significant foreground region in an image, thereby guiding network learning to enhance the global contrast. The invention has the beneficial effects that the performance is better, and meanwhile, the UIQM score is better.

The invention discloses a multi-lens stereoscopic vision parallax calculating method, and belongs to the field of computer vision. The method comprises four steps: initialization, the obtaining of all binocular parallaxes for range finding, the obtaining of a multi-lens parallax for range finding, and ending. A weighted averaging method is employed for the calculation of a final parallax, and can eliminate a range finding error better. A scene is observed from a plurality of viewpoints, so as to obtain perception images at different visual angles. The position deviation among pixels of the images is calculated through a triangular measurement principle, so as to obtain the three-dimensional information of the scene. The multi-lens stereoscopic vision parallax method is given on the basis of a stereoscopic parallel vision model. A parallax calculation method of the stereoscopic parallel vision model is employed for the mutual parallax calculation of a plurality of lenses, thereby obtaining a more precise imaging parallax, and enabling the depth information of object points in three-dimensional reconstruction to be more precise. The plurality of lenses can rotate around a circle center in each distribution mode at a proportion with the unchanged relative distance, or can be arranged in a turnover manner along the axis of a line in a plane.

The invention discloses a compression perceptual image reconstruction method based on multi-view images. Firstly, for each target reconstruction, a corresponding dynamic image set is selected by calculating a parallax compensation size between images, and then multi-view image reconstruction is realized based on the selected dynamic image set. In the reconstruction process, high-quality reconstruction results are obtained according to the adaptive total variational regularization constraint based on disparity compensation and non-local low-rank tensor constraint.

The invention discloses a product perceptual image vocabulary extraction method and system, and the method comprises the following steps: collecting comment text data of a target product, carrying outthe word segmentation of the comment text data, and obtaining evaluation vocabularies; extracting high-frequency vocabularies for evaluating the appearance in the evaluation vocabularies as central vocabularies, and extracting adjectives in the evaluation vocabularies to obtain adjective vocabularies; converting the evaluation vocabulary into a word vector, calculating the similarity between eachadjective vocabulary and the central vocabulary based on the word vector, and extracting the corresponding adjective vocabulary as an initial perceptual image vocabulary according to the similarity;and clustering the initial perceptual image vocabularies, and extracting the corresponding initial perceptual image vocabularies as perceptual image vocabularies according to a clustering result. Theperceptual image vocabularies of the target product can be objectively and accurately extracted based on the comment text data, meanwhile, the labor cost can be reduced, and the extraction efficiencyis improved.

The invention discloses a joint task learning method for super-resolution and perception image enhancement of a single image, and the method comprises usually carrying out the hybrid combination of asuper-resolution task and a perception image enhancement task for the actual demands in an actual scene, and obtaining a high-quality high-resolution enhanced image from a low-resolution original image. The invention provides a super resolution and perception image enhancement task joint learning framework named Dep SR-PIE. The framework comprises a multi-path super-resolution network (MSRnet), adetail complementary network (DCN) and a hybrid U-net enhancement network (FULENet). The MSRnet utilizes a multipath learning strategy to describe local and global information at the same time, the DCN utilizes double bypass shared convolution to sample and enhance high-frequency details, and the FULENet seeks an optimal fusion color correction matrix to learn color and tone mapping. Through quantitative and qualitative evaluation of the four data sets, a conclusion that most indexes of a joint learning framework are superior to those of a comparison method can be obtained. Through the methodprovided by the invention, a high-quality high-resolution enhanced image can be obtained more quickly and efficiently.