84 results about "Human visual perception" patented technology

The invention discloses a super-resolution image reconstruction method based on a multi-scale large-kernel convolution double-residual neural network, which is suitable for the field of image processing, and comprises the following steps: cutting a data set, inputting a cut original low-resolution image into a preprocessing module, carrying out image normalization and data enhancement operation, and carrying out image reconstruction; generating a preprocessed low-resolution image; the preprocessed low-resolution images form a distorted image block data set, and a training set, a verification set and a test set are formed; according to an existing distorted image block data set, a super-resolution image reconstruction method based on a multi-scale large-kernel convolution double-residual neural network is constructed; and inputting the data set into the constructed multi-scale large-kernel convolution double-residual neural network to extract semantic features, and amplifying a feature map by using an up-sampling module of the model to generate a super-resolution image. According to the method, a multi-scale large-kernel convolution and double-residual structure is introduced, a visual attention mechanism is used in the neural network, the extracted features better conform to human visual perception features, and super-resolution image reconstruction is more accurate.

The invention discloses a non-reference image quality evaluation method and device based on multi-perception feature fusion and dynamic enhancement, and the method comprises the steps: obtaining the quality information of a multi-domain distorted image based on superpixel segmentation and Gaussian kernel filtering texture generation; quality related feature extraction is performed on multi-domain information through a semantic perception module and a distortion perception module, bidirectional modulation is performed on global visual features of an original distorted image through a cross attention mechanism, and dynamic fusion of multiple perception features is realized; through a parallel feature enhancement unit formed by local adaptive filtering of a visual self-attention block and a dynamic residual block, dynamic allocation of perception modes to different content areas is realized; generating a weighted quality score consistent with human visual perception through a weighted dual-path regression device; and outputting a predicted score consistent with the human score from the distorted image through the three sub-networks. According to the method, the problems of insufficient adaptability to complex content of a distorted image and low local distortion sensitivity are effectively solved.

The invention relates to an electronic paper display mapping method based on visual perception and dynamic clustering, and belongs to the technical field of display. According to the method, firstly, learning factors and particle weights of a particle swarm optimization algorithm are dynamically adjusted, a gray segmentation threshold is optimized in combination with K-means clustering, and the accuracy and convergence efficiency of gray distribution are remarkably improved; secondly, establishing a dynamic error diffusion system based on visual perception, designing a frequency domain visual weighted filter and a spatial domain sensitivity function by combining a human visual perception theory, dynamically compensating quantization errors and enhancing edge details; and meanwhile, a Floyd-Steinberg error diffusion algorithm is improved, and a double image effect is inhibited in combination with a snakelike scanning path. According to the electronic paper display mapping method based on visual perception and dynamic clustering, the electronic paper image display quality is effectively improved, particularly when a high-gray-scale image is displayed, details can be reserved to the maximum extent, and the problems of gray-scale distortion and edge blurring in traditional electronic paper display are solved.

The invention discloses an event-visible light combined video enhancement method for cross-modal spatial-temporal correlation analysis, and the method comprises the steps: 1), generating a corresponding dynamic event stream frame through an event camera simulation model, and dividing a data set; (2) a Feature Extractor module based on UNet is constructed, and feature extraction is carried out; 3) constructing a cross-modal reconstruction unit based on BCPS, and recovering contour information of the visible light video frame; 4) constructing a space domain noise filtering network VSNF and a time domain noise filtering network VTNF, and performing feature analysis respectively; 5) constructing a UNet-based feature fusion module, and generating a denoising result conforming to human visual perception characteristics; and 6) designing a loss function, and achieving an optimal balance state through multiple times of loop training. The method belongs to the technical field of image processing, and solves the problems of incomplete dynamic feature extraction, insufficient time domain information utilization and low overall visual quality of a video in a denoising process in the prior art.

The invention provides a dynamic feature contribution degree-based color textile fabric image retrieval method. Innovation and improvement are carried out aiming at single feature limitation and fixed weight fusion defects existing in a traditional retrieval technology. Color, texture and shape features are separated through a multi-modal feature decoupling technology, and coupling interference between the features is eliminated; a dynamic contribution degree quantification mechanism is innovatively constructed, gradient back propagation is used for calculating the contribution proportion of features to retrieval results in real time, and weight self-adaptive adjustment is achieved. According to the method, the problem of retrieval deviation caused by complex characteristics such as color gradient and blending structure difference of the colored spun yarn fabric is solved in a breakthrough mode, weight distribution can be intelligently matched according to inquired image characteristics (for example, the color gradient fabric focuses on color characteristics, and the jacquard fabric focuses on shape edges), and the retrieval result better fits human visual perception. According to the technical scheme, the image retrieval precision of complex fabrics in the textile industry is effectively improved, and an intelligent solution is provided for production process data association.

The invention relates to the technical field of digital image processing, and discloses an image defogging method based on an atmospheric scattering model and color correction, and the method comprises the following steps: S1, carrying out the color correction of a foggy image with color cast; s2, calculating an atmospheric light value of the foggy image; s3, calculating the transmissivity value of the foggy image; s4, recovering the defogged image by adopting an atmospheric scattering model according to the atmospheric light value and the transmissivity value; and S5, solving a defogged image after adaptive brightness enhancement and adaptive contrast enhancement. According to the method, the color of the obtained defogged image can be recovered to be natural, the phenomena of color cast, incomplete defogging, detail information loss and overall darkness of the image are improved, the recovered defogged image better conforms to the perception of human eyes, meanwhile, the implementation process is simple, the time complexity is low, and better defogging efficiency is achieved.

The invention discloses a super-resolution image reconstruction method based on a multi-scale large separable kernel convolutional neural network, which is suitable for the field of image processing, and comprises the following steps: cutting a data set, inputting a cut original low-resolution image into a preprocessing module, carrying out image normalization and data enhancement operation, and carrying out image reconstruction; generating a preprocessed low-resolution image; the preprocessed low-resolution images form a distorted image block data set, and a training set, a verification set and a test set are formed; constructing a super-resolution image reconstruction method based on a multi-scale large separable kernel convolutional neural network according to an existing distorted image block data set; and inputting the data set into the constructed multi-scale large separable kernel convolutional neural network to extract semantic features, and amplifying a feature map by using an up-sampling module of the model to generate a super-resolution image. According to the method, a multi-scale large separable kernel convolution structure is introduced, a visual attention mechanism is used in the neural network, the extracted features better conform to human visual perception features, and super-resolution image reconstruction is more accurate.

The invention provides a robust anti-attack method for a target detection system based on a reference image. The method comprises the following steps: selecting a detector, selecting a pedestrian image data set with a real label as a training data set, and training the detector by using the training data set to obtain a trained detector; selecting a reference image as an initialized patch, performing enhancement transformation processing on the initialized patch to obtain an optimized patch, generating an adversarial sample by using the optimized patch, inputting the adversarial sample into the trained detector to obtain a detection result, and further optimizing the patch through back propagation to obtain a final patch; and utilizing the final patch to carry out countermeasure attack on a detector in target detection. According to the method, the problems of non-concealment, poor robustness and the like of an adversarial patch in an existing scheme are solved, a novel adversarial patch method which is natural in appearance and conforms to human visual perception is provided, and meanwhile, the robustness of the patch in environment change and multi-view scenes can be improved.

The invention discloses a self-adaptive multi-depth-of-field sub-pixel layer defect detection method, which comprises the following steps of: acquiring pictures of a sub-pixel layer through an image acquisition module to obtain different depth-of-field images; the collected image is preprocessed based on an image preprocessing module, specifically, multiple filters are adopted to carry out noise reduction and smoothing processing on the input image, so that noise interference is eliminated, and the image quality is improved; a defect detection module is adopted to accurately extract edge features in images with different depth of field; calculating multi-dimensional feature information of the image through a feature calculation module; and designing defect evaluation indexes of human eye visual perception based on the obtained multi-dimensional feature information, and accurately positioning the clearest layer of the defect by adaptively adjusting the weight of each evaluation index and combining the depth information of the multi-depth-of-field image so as to determine the optimal distribution position of the defect in the sub-pixel layer. Meanwhile, a contrastive analysis mechanism consistent with human visual perception is introduced, and it is ensured that the detection result is highly consistent with human visual perception.

The invention provides a remote sensing image self-learning segmentation method, belongs to the technical field of image processing, and aims to realize automatic, accurate, sufficient and reliable high-resolution remote sensing image segmentation. Comprising the following steps: responding to an input remote sensing image, and performing spectrum correction operation on the remote sensing image to obtain a first color image; carrying out super-pixel division on the first color image by utilizing an edge recognition and super-pixel division method for simulating a non-classical receptive field to obtain super-pixel plaques under different scales; performing clustering operation on the superpixel plaques to obtain a clustering result; training a multi-feature deformation lightweight neural network based on a visual attention mechanism by using a clustering result, and generating a super-pixel region recognition model; and performing remote sensing image segmentation by using the super-pixel region recognition model. According to the method, the self-learning ability of human eye visual perception can be simulated, then image segmentation is carried out through self-adaptive analysis, self-learning identification and self-checking correction, and a segmentation result is rapidly and accurately obtained.

The invention discloses an infrared-visible light image fusion method based on frequency domain high-order interaction. The infrared-visible light image fusion method comprises the following steps: acquiring an infrared image and a visible light image; respectively extracting initial features of the infrared image and the visible light image through a cross center difference convolution operator, inputting the initial features into an FDB module to obtain frequency domain primary features, and inputting the initial features of the infrared image into an ATB module to obtain a weight matrix; using Hadamard product processing to obtain visible light frequency domain features, and gradually refining the initial features of the infrared image through convolution under the guidance of a weight matrix to obtain infrared frequency domain features; inputting the obtained visible light and infrared frequency domain features into a channel interaction module to obtain primary visible light features; and aggregating the primary visible light feature output and the initial feature of the infrared image, outputting the aggregated primary visible light feature output and the initial feature of the infrared image as the input of the FDB module of the next layer, iterating the process, and outputting the final fusion feature. The method is used for generating a fused image which has fine texture and higher contrast and better conforms to human visual perception.

The present invention discloses a remote sensing image scene boundary recognition method that conforms to human visual perception. A remote sensing raster image is input, and the image is segmented at the object level and the scene level respectively to obtain the corresponding object spot raster image and the initial scene spot raster image, and then the initial scene spot is updated with the object spot to obtain the final scene spot raster image, and finally the final scene spot raster image is subjected to a raster-to-vector operation to obtain the remote sensing image scene boundary that conforms to human visual perception. The method is simple in principle, easy to implement, and highly automated, and has a wide range of application value in the fields of remote sensing and geography, and is not only helpful in identifying suspicious targets in various scenes of remote sensing images, but also helpful in determining ecological landscape boundaries, regional planning boundaries, and natural geographical zoning boundaries.

The invention discloses a quantitative detection method for anti-perspective performance of a textile in a complex light environment, and belongs to the technical field of new material detection, a micro gap is arranged between the textile to be detected and a variable-frequency gray scale target to simulate a non-clinging state when clothes are worn and excite Fresnel diffraction and subsurface scattering effects, and the anti-perspective performance of the textile is detected. Complex illumination conditions such as backlight and sidelight are simulated by adjusting the light intensity ratio, and after a target perspective image is acquired by using an image acquisition unit, the modulation transfer function attenuation rate of a to-be-detected textile to the target is calculated, so that the anti-perspective performance of the textile is quantitatively represented; the interference of glare on the surface of the high-gloss fabric is effectively eliminated by introducing an orthogonal polarization modulation technology, and a detection result is highly consistent with the visual perception of human eyes by adopting a specific light source spectrum simulating the skin color of a human body; compared with a traditional light transmittance method, the method has the advantages that the dimension crossing from luminous flux measurement to image information fidelity evaluation is realized, and the accuracy, reliability and practicability of detection are remarkably improved.

The application discloses an infrared and visible light image fusion algorithm based on a self-encoder, and comprises the following steps: S1, an encoder network is constructed by combining a short cross-layer connection with an ECA attention mechanism; S2, registered infrared and visible light images are input into the encoder network, and the encoder and decoder network are trained to obtain an encoder meeting a condition; S3, feature fusion is performed on the input images by using the constructed fusion network, and the fusion network is trained to obtain a fusion network meeting a condition; and S4, the fused images are decoded and reconstructed by using the decoder.The evaluation index of the fused images is significantly improved, the target is clear, the details are prominent, the outline is obvious, and the images meet the human visual perception.

The application discloses a method for testing response time of a mini / micro LED direct display screen, which can avoid the influence of subjective evaluation of human visual perception on the response time, and provides a means for high-precision testing of the response time of the direct display screen through quantitative evaluation method. The method comprises the following steps: (1) a power supply inputs a stable voltage to the direct display screen, the direct display screen is lighted when a square wave driving voltage is input, light enters a microscope lens and converges on a target surface of a high-speed photoelectric detector, a response voltage signal is generated after the target surface is irradiated by the light, the voltage signal is filtered by a high-frequency filter to suppress high-frequency noise, and the voltage signal is connected to an oscilloscope through a BNC line; (2) the power supply is driven and controlled to realize different refresh rates; and step (1) is repeated to realize the response time test of the direct display screen under different refresh rates; and (3) different colors are realized by controlling a signal of the driving power supply; and step (1) is repeated to realize the response time test of the direct display screen under different colors.

The application discloses a perceptual-geometry joint code rate control method, system and medium, and belongs to the field of video coding. The method comprises the following steps: assigning weights to each CTU in each frame in a video according to the VVC standard; for each I frame, calculating the perceptual-geometry joint weight of each CTU in the I frame, and updating the weight of each CTU in the I frame; and assigning a code rate to each CTU according to the weight of each CTU in each frame to complete code rate control. The application can effectively eliminate geometric redundancy and improve human visual perception consistency, and promote efficient transmission of high-resolution videos.

The invention provides an infrared and visible light image fusion method based on a second-order attention mixed feature, and the method comprises the steps: inputting the first-order and second-order statistics of an infrared image and a visible light image, and adaptively integrating the global features and local features of the infrared and visible light images, thereby enabling the fused image to obtain a stronger feature expression, and improving the image fusion precision. Pixel intensity and texture details are reserved at the same time, comparable contrast is kept while a more vivid edge is achieved, an output result keeps rich edge and texture information transmitted by an input image, artifacts are few, and visual perception of human beings is better met. Compared with other infrared and visible light image fusion algorithms based on deep learning, the method has a better fusion effect.

This invention discloses a method for fusing infrared and visible light images based on multi-layer convolution. The network structure includes an encoder network, a decoder network, and a multi-layer convolutional fusion network. The encoder is composed of nested multi-layer convolutional blocks and an ECA attention mechanism. The decoder is mainly composed of decoding blocks, each of which consists of two convolutional layers. The multi-layer convolutional fusion network is mainly composed of gradient convolutional blocks, downsampling convolutional blocks, a convolutional spatial channel attention mechanism, and several convolutional layers. The method includes the following steps: S1, the registered infrared source image and visible light source image are fed into the encoder in pairs, and the encoder extracts the source image features; S2, the multi-layer convolutional fusion network fuses the source image features to obtain the fused features; S3, the decoder reconstructs the fused features and outputs the image. The fused image produced by this invention has prominent targets, clear details, obvious contours, and significantly improved performance indicators, conforming to human visual perception.

This application relates to the field of image processing technology, and provides a color compensation system and method based on adaptive color adjustment. Through the close combination of four steps, from the acquisition of multi-source data, the determination of key parameters, the output of the compensation parameter matrix to the final compensation optimization, each step progresses layer by layer and supports each other, which can effectively improve the color restoration degree of images under different ambient lights and target device conditions, ensure that the output RGB images have accurate and natural colors, conform to human visual perception, and significantly improve the quality and efficiency of image color processing, providing strong technical support for various application scenarios involving image display and processing.

The application provides a color spun fabric image retrieval method based on dynamic feature contribution degree, and innovatively improves the single feature limitation and fixed weight fusion defects existing in traditional retrieval technology. Color, texture and shape features are separated through a multi-modal feature decoupling technology to eliminate the coupling interference between features; a dynamic contribution quantification mechanism is innovatively constructed, the contribution proportion of features to retrieval results is calculated in real time by using gradient back propagation, and adaptive adjustment of the weight is realized. The method breaks through the retrieval deviation problem of color spun yarn fabrics caused by complex characteristics such as color gradient and difference in blended structure, can intelligently match the weight distribution according to the query image features (such as color gradient fabric focusing on color features and jacquard fabric focusing on shape edges), and makes the retrieval results more consistent with human visual perception. The technical scheme effectively improves the image retrieval accuracy of complex fabrics in the textile industry and provides an intelligent solution for production process data association.

The application discloses a kind of based on layered feature fusion's no reference image usability evaluation method.There is following step: first, respectively extracting the low-level picture quality feature vector of the image to be evaluated, middle layer structure feature vector and high-level semantic feature vector;Subsequently, respectively to feature vector is normalized and splicing;Then, total feature vector is input to the regression network of integrated layered attention fusion ware, the network passes through multi-head self-attention mechanism dynamically learns the contribution weight of each level feature to the final availability score, and carries out regression prediction, outputs the availability evaluation score of image.The application realizes the fundamental change of evaluation target from "human visual perception quality" to "machine task availability" by systematically fusing multidimensional features from picture quality to semantics, and the evaluation result is strongly related to the performance of high-level visual tasks such as object detection and pose estimation, and is suitable for image screening and quality control in scenarios such as autonomous driving and intelligent monitoring.

The invention discloses a lamp light quality detection system and method, relates to the technical field of lamp detection, and solves the problems that traditional detection depends on physical parameters and cannot reflect real visual perception of a human body, and the system comprises a spectacle frame, and a physiological signal sensor, a signal processing unit and a light quality detection unit which are integrated on the spectacle frame. Visual physiological reaction signals of a user under illumination of the lamp are collected in real time through the sensor, and after physiological features are extracted through the processing unit, quantitative detection indexes of light quality are output through a model built in the detection unit. According to the invention, objective and personalized in-situ detection and evaluation of the light quality of the lamp in a natural eye using state are realized, and the method is mainly suitable for field detection and health evaluation of various lighting environments such as classrooms and offices.

The invention discloses a physical perception fusion driven adaptive texture generation optimization method and system. The method comprises the following steps: collecting spectral reflectivity data of a material and microscopic texture image data of the surface of the material, and preprocessing the spectral reflectivity data and the microscopic texture image data; constructing an initial physical and perception fusion driven texture generation model; carrying out training optimization on the model by adopting a cross-domain joint training strategy through a physical priority-perception priority-joint fine tuning alternative optimization scheme to obtain a physical and perception fusion driven texture generation model; and performing actual texture generation based on the obtained physical and perception fusion driven texture generation model. According to the method, a new-generation adaptive texture optimization framework is constructed by deeply fusing a physical rule and a human visual perception mechanism, the problem of physical-perception splitting existing in the high-fidelity rendering field for a long time is solved, and breakthrough in spectrum modeling precision, rendering efficiency and visual quality is achieved.

The invention discloses an intelligent ship sensing method based on scene complexity. The method comprises the following steps: defining a scene complexity level; screening image texture feature quantities; calculating image texture feature parameters; classifying scene complexity by using an ensemble learning network model; according to the method, uncertain factors in ship navigation scene images are comprehensively considered, image features are extracted by using a gray-level co-occurrence matrix, and an XGBoost network model of integrated learning is used for sensing the complexity of a navigation scene. The method comprises the following steps: by simulating human visual perception, combining a plurality of parameters such as energy, entropy, contrast, inverse difference moment and correlation of an image, and constructing a mathematical model capable of truly reflecting scene complexity; the method provides important reference for the design and construction of the autonomous navigation scene of the intelligent ship, and helps to improve the safety and reliability of the intelligent ship in actual navigation by evaluating the complexity of the virtual test scene.

This invention relates to the field of HDR reconstruction, specifically to the task of reconstructing HDR images from multi-exposure LDR images. A perceptual artifact detection and elimination method for high dynamic range reconstruction is proposed, comprising the following steps: Step 1, establishing an HDR-AD dataset; Step 2, constructing and training an HDR-ADet model; building an artifact detector HDR-ADet based on ViT, and training this detector using the HDR-AD dataset to locate artifact regions in the image; Step 3, using the artifact mask obtained from HDR-ADet to fine-tune the HDR reconstruction model to improve the quality of the reconstructed HDR image; and further improving the quality of the reconstructed HDR image by integrating HDR-ADet into the loss function calculation process of the HDR reconstruction framework. This invention can improve the quality of HDR imaging, achieving breakthroughs not only in quantitative indicators but also more closely approximating human visual perception.

The invention relates to an evaluation method and system applied to restored images of ancient Chinese calligraphy and painting, and the method comprises the steps: 1, taking the restored images of a first ancient calligraphy and painting and a second ancient calligraphy and painting, which are restored through an artificial intelligence technology and PS software, as evaluation preprocessing samples, selecting a peak signal-to-noise ratio, a structural similarity index and a color difference to combine with a subjective visual MOS score to evaluate a restoration result; 2, according to the evaluation condition, optimizing the weight of each feature component in the ancient calligraphy and painting repair evaluation for one objective evaluation index in the step 1, and the like. The evaluation system comprises an ancient calligraphy and painting repairing module, an evaluation module and a verification module. The invention further provides electronic equipment and a computer readable storage medium which are applied to the evaluation method and system. The method has the advantages that the restoration result of the ancient calligraphy and painting is assessed through a multi-dimensional assessment system, four indexes of PSNR, SSIM, HSV and MOS are fused, and the texture, color, structure and human eye perception full-dimensional requirements of restoration of the ancient calligraphy and painting are covered.

This disclosure presents an efficient rendering method based on scene-aware mapping for complex scenes. One specific implementation of this method includes acquiring a scene model, camera parameters, and user gaze point information corresponding to the target scene; performing spatial mapping processing on the scene model based on the camera parameters to obtain a two-dimensional geometric scene graph; generating a consistent visual importance graph based on the user gaze point information and the two-dimensional geometric scene graph; generating a pixel size control graph based on the two-dimensional geometric scene graph and the consistent visual importance graph; and rendering the two-dimensional geometric scene graph based on the pixel size control graph to obtain a rendered scene image. This implementation can reduce the computational resources required for rendering and improve rendering speed while ensuring the quality of human visual perception.

The application discloses a blind image quality evaluation method based on multi-scale features and long-distance dependence, and comprises the following steps: step 1, acquiring a training sample set and a test sample set; step 2, constructing a network model of a no-reference image quality evaluation method based on multi-scale features and long-distance dependence; the network model is used for extracting and fusing multi-scale features and long-distance dependence of an image to regress to an image quality score; step 3, iteratively training the network model of the no-reference image quality evaluation method based on multi-scale features and long-distance dependence; and step 4, acquiring a no-reference quality evaluation result of the image. The application is used for solving the problems of large parameter quantity and low calculation efficiency of an existing method and the problem that the existing method neglects the fusion of long-distance dependence between multi-scale features and local quality features, thereby leading to low consistency between a model prediction result and human visual perception.

The application discloses a GAN image repairing method based on stable field jump connection, embeds a stable field operator into a jump connection, uses the stable field operator to preliminarily predict damaged pixels of an encoder feature map in a generation unit, then transfers to a decoder and outputs a generated image. Then, the type of loss is refined, the loss function is redefined as an adversarial loss, a pixel reconstruction loss and a pyramid loss to further guide the training of the network to the correct direction. The method improves the effect of damaged image repairing, meets the demand of human visual perception, and is similar to the original image to a certain extent. And experiments prove that the method achieves good repairing effect on face images, natural scene images and building images.

The invention discloses a multispectral optical signal acquisition and AI intelligent analysis device and method, and belongs to the technical field of optical signal detection and artificial intelligence. A multi-spectral optical signal intelligent analysis deep network system (MSS-IDNS) is defined as a core, and the system takes a 300-1000nm multi-spectral optical signal as an input carrier and can be fused with various deep learning models. The device comprises a multispectral optical signal acquisition terminal and an AI intelligent analysis module. The method comprises the six steps of optical signal collection, data preprocessing, feature extraction, AI analysis, model iteration and intelligent response, so that intelligent analysis and biological state recognition of optical signals are completed. The method breaks through the human visual perception boundary and adapts to multiple scenes, core hardware is an existing mature industrial product, batch production can be achieved, and the method has extremely high civilian mass production potential and commercial value.