Patents
Literature
Hiro is an intelligent assistant for R&D personnel, combined with Patent DNA, to facilitate innovative research.
Hiro

38results about How to "Rich detail features" patented technology

Image processing method and device, storage medium and mobile terminal

The invention discloses an image processing method, an image processing device, a storage medium and a mobile terminal. The image processing method is applied to the mobile terminal and comprises thesteps of: acquiring an original image collected by the mobile terminal, wherein the color depth of the original image is at least 10 bits per pixel; determining a brightest region in the original image, and performing exposure processing on the brightest region to obtain intermediate image data; and performing logarithm conversion on the intermediate image data to obtain a target image. Accordingto the image processing method of the invention, gamma correction does not need to be carried out on the intermediate image data, but logarithm conversion is carried out, and more detail features of bright parts and dark parts of the image are reserved by using logarithm conversion; and on the other hand, exposure processing is carried out according to the brightest region, detail feature loss isavoided, and more detail features are reserved.
Owner:HUIZHOU TCL MOBILE COMM CO LTD

Three-dimensional face reconstruction method and device, electronic equipment and storage medium

The invention discloses a three-dimensional face reconstruction method and device, electronic equipment and a storage medium, and relates to the technical field of electronic equipment. The method comprises the following steps: acquiring a shape parameter and a texture parameter of a to-be-reconstructed face, inputting the shape parameter into a first model to obtain a three-dimensional face shape output by the first model, and inputting the texture parameter into a second model to obtain a face texture map output by the second model, wherein, at least one of the first model and the second model is obtained based on generative adversarial network training, a target three-dimensional face is generated based on the three-dimensional face shape and the face texture map, and the target three-dimensional face comprises texture information generated based on the face texture map. According to the method, the three-dimensional face shape and / or the face texture map are / is generated through the trained generative adversarial network, so that the generated target three-dimensional face has rich detail features, and the reconstruction effect of the three-dimensional face is improved.
Owner:GUANGDONG OPPO MOBILE TELECOMM CORP LTD

Single-frame image three-dimensional reconstruction device and method based on deep learning

ActiveCN107862741ASolve the problem of high-precision 3D reconstructionRich detail featuresDetails involving processing stepsImage enhancementNetwork modelStudy methods
The invention discloses a single-frame image three-dimensional reconstruction device and method based on deep learning. The device comprises an upper computer, a support frame and a high-definition camera at the top thereof. Three parallel light LED area light sources in 120 degrees are installed in equal height around the high-definition camera. The method comprises the following steps: shootinga training sample, using images simultaneously irradiated by red green blue lights as input data needed by a training model, taking a gray value of three images successively irradiated by the white light, turning into the single-channel image as a truth-value for the training model, using a deep learning method and the input data, constructing a full linked network model of per-pixels for single-frame image three-dimensional reconstruction, training the model, continuously regulating and optimizing network parameters by using a counterpropagation algorithm, forecasting target surface three-dimensional information, and three-dimensionally re-constructing the result of the network forecast by using a luminosity three-dimensional algorithm, to obtain the object surface three-dimensional information. The method is capable of, through improving the network structure, constructing the network model for the single-frame image three-dimensional reconstruction, and increasing the application scene compared with the multi-frame image three-dimensional reconstruction.
Owner:中译文娱科技(青岛)有限公司

Vehicle type fine identification method and system

The invention discloses a vehicle type fine identification method and system. The method comprises the following steps: carrying out graying and standardizing processing on an obtained original vehicle image to obtain a standardized image; calculating gradient and direction of each pixel point of the standardized image; carrying out direction gradient histogram feature extraction and local linear constraint coding on the standardized image according to the calculated gradient and direction to obtain encoding vector of the standardized image; processing the standardized image obtained after local linear constraint coding through a weight space pyramid according to the obtained encoding vector to obtain final expression vectors of the vehicle image, wherein the final expression vectors of the vehicle image comprise position information and semantic information of the vehicle image; and inputting the final expression vectors of the vehicle image to a pre-trained linear support vector machine classifier for vehicle type identification. The vehicle type fine identification method and system have the advantages of high accuracy, low complexity, high robustness and rich detail features, and can be widely applied to the field of picture processing.
Owner:SUN YAT SEN UNIV +1

A smoke detection method based on multi-network model fusion

The invention relates to a smoke detection method based on multi-network model fusion, and two network models of VGG16 and ResNet50 are fused to realize reliable detection of smoke. According to the fusion network provided by the invention, more abundant smoke image detail features can be extracted, and the distinguishing capability of the features on the smoke image and the smoke-like image is enhanced. By adopting a feature transfer learning method based on an isomorphic space, feature extraction layers of pre-trained VGG16 and ResNet50 models can be well migrated to a target data set classification task in a smoke scene, and meanwhile, the generalization capability of the model is improved. By fusing the multi-network model, the distinguishing capability of the characteristics is enhanced, the false alarm phenomenon caused by targets such as cloud and fog similar to smoke is reduced, and the reliability of smoke detection is further improved.
Owner:NAT UNIV OF DEFENSE TECH

Real-time depth completion method based on pseudo depth map guidance

The invention discloses a real-time depth completion method based on pseudo depth map guidance. The method includes: enabling an RGB camera to collect an RGB image, and enabling a laser radar to detect to obtain a sparse depth image; performing data processing of morphological operation on the sparse depth map to generate a pseudo depth map; establishing a neural network structure, and processing the pseudo depth map, the RGB map and the sparse depth map to obtain a dense depth map; training the neural network structure, setting total loss function optimization supervision, and obtaining parameter values of all parameters in the neural network structure; and loading the parameter values of all parameters into the neural network structure, inputting a single-frame sparse depth map to be measured into the neural network structure, and outputting a dense depth map. According to the method, the precision of depth completion is effectively improved, wrong pixels of a sparse depth map can be corrected, structural similarity supervision is provided, the depth residual image is regressed during prediction, and a high-precision depth completion result can be obtained under a high real-time condition.
Owner:ZHEJIANG UNIV

Image processing method and system based on image generation network model, and storage medium

The invention relates to an image processing method and system based on an image generation network model, and a storage medium. The method comprises the following steps: pre-constructing an image generation network model comprising a perception up-sampling convolution module; sensing an up-sampling convolution module by using the feature map obtained by down-sampling; enabling first convolution channel to perform convolution of a first scale on the feature map to generate a first convolution feature map; enabling a second convolution channel to perform second-scale convolution on the featuremap to generate a second convolution feature map; enabling a fusion layer to perform fusion connection on the first convolution feature map and the second convolution feature map to generate a fusionfeature map; enabling a sub-pixel conversion layer to convert the fusion feature map into an output picture through a depth-to-space function, wherein the size of the output picture is twice that of the feature map. According to the image processing method provided by the invention, in the up-sampling process, convolution operation of two different sizes is carried out to acquire more detail features, the feature characterization of the output picture is improved, and the chessboard effect is improved.
Owner:SHENZHEN UNIV

Optimization method of HarDNet-Lite in embedded platform

The invention discloses a method for optimizing HarDNet-Lite on an embedded platform, which is used for solving the problems that the existing target detection network is too complex, the operand is large, the reasoning speed on the embedded platform is low, and the positioning precision is low. The method comprises the following steps: 1) establishing a lightweight HarDNetLite feature extractionnetwork; 2) fusing feature maps of different scales by adopting a weighted FPN structure, so rich underlying detail information and high-level semantic information are fully fused; 3) generating a YOLO detection head, and placing an anchor frame generated through k-means clustering on the feature maps of different sizes to detect targets of different sizes; 4) performing end-to-end model trainingby using a classification and regression loss function; 5) deploying the trained model on an embedded platform, and carrying out target detection. The invention has the beneficial effects that the reasoning speed and the target detection precision of the HarDNet-Lite target detection network on the embedded platform are improved.
Owner:苏州凌图科技有限公司

Image super-resolution optimization method based on similar image retrieval, medium and equipment

The invention discloses an image super-resolution optimization method based on similar image retrieval, a medium and equipment, and belongs to the field of image data processing. The method comprises the steps that firstly, similar high-resolution images are retrieved from a high-resolution image material library; then extracting features of the similar high-resolution images, introducing an attention mechanism, taking a high-resolution intermediate image obtained by a bicubic interpolation algorithm as a query, extracting features related to the low-resolution images in the similar high-resolution images, and fusing the features related to the low-resolution images with the features of the low-resolution images after up-sampling to generate high-resolution images; and finally, taking the generated high-resolution image as a new query of the attention mechanism, and obtaining a final high-resolution image generation result through iteration. According to the method, related detail features of the high-resolution image similar to the input image are combined, feature expression of the low-resolution image is enhanced through an attention mechanism, and a high-resolution version of the low-resolution image is better generated.
Owner:杭州碧游信息技术有限公司

Living fingerprint recognition device

The invention discloses a living fingerprint recognition device comprising a microcontroller and a phased array biological recognition module, a detection module, a storage and a power supply module connected with the microcontroller, wherein the microcontroller is used for processing information of the phased array biological recognition module and the detection module in real time; the phased array biological recognition module is used for recognizing a fingerprint; the detection module is used for collecting physiological feature information of a user; the storage is used for storing the fingerprint information and the physiological feature information of the user; and the power supply module powers the microcontroller, the phased array biological recognition module, the detection module and the storage. The device provided by the invention applies a phased array principle to a fingerprint sensor, a recognition rate and accuracy of the fingerprint sensor are improved, the risk that the fingerprint recognition device is cheated by the counterfeit false fingerprint is reduced by adding the detection module, and security of the fingerprint recognition device is improved.
Owner:XINFOO SENSOR TECH CO LTD

Video detection and comparison system

The invention belongs to the technical field of video comparison, and particularly relates to a video detection and comparison system, which comprises (1) a data acquisition module; (2) a video fingerprint synthesis module; (3) a first verification module; and (4) a verification module II. According to the invention, feature extraction is carried out on each key frame in the target video; the extracted image features of each key frame are directly taken as a target video fingerprint; satellite three-dimensional data and a user actual three-dimensional data reconstruction picture are introducedto construct a video scene live-action model to obtain a reconstructed video key frame; and the similarity between the target video and the reference video is determined through mutual verification of the obtained reconstructed video key frame and the key frame number distance data of the verification module I. The outdoor video comparison accuracy is further improved, repeated comparison of videos is reduced through comparison, and time is saved.
Owner:GUIZHOU YONGXING TECH

Image processing method and device, electronic equipment and computer readable storage medium

The embodiment of the invention provides an image processing method and device, electronic equipment and a computer readable storage medium, and belongs to the field of image processing, and the method comprises the steps of obtaining at least two feature maps with different spatial resolutions based on a to-be-processed image; and carrying out at least one wavelet analysis based on the at least two feature maps to obtain an up-sampling feature map with the target spatial resolution. Based on the scheme provided by the embodiment of the invention, the up-sampling feature map can be obtained by combining the features of the at least two feature maps with different spatial resolutions, so that the obtained up-sampling feature map has more detail features, and the image is clearer.
Owner:BEIJING SAMSUNG TELECOM R&D CENT +1

Medium-long wave dual-band infrared image fusion method

The invention discloses a medium-long wave dual-band infrared image fusion method. The method comprises the following steps: 1, ; reading an image to be fused, processing a Y channel, extracting detail features of a medium-wave image, extracting difference features of the medium-wave image and a long-wave image, calculating reference mean values of a Cb channel and a Cr channel, calculating reference variances of the Cb channel and the Cr channel, calculating a mean value and a variance of a source image, transferring colors, superposing the features, and outputting a final fused image. The medium-long wave infrared image fusion algorithm provided by the invention is based on a color transfer theory, does not need to refer to that the image has a good visual effect in each scene, has extremely high environmental adaptability, has more radiation characteristics and detail characteristics, and is easy to popularize and apply.
Owner:KUNMING INST OF PHYSICS

Highlight image enhancement method and device based on dark channel prior mirror surface and storage medium

ActiveCN111968062AEnhance Specular Highlight ImageRestore local informationImage enhancementImage analysisContrast enhancementSpecular highlight
The invention discloses a mirror surface highlight image enhancement method and device based on dark channel prior, and a storage medium. The method comprises the steps: selecting the most fuzzy pixelin an input image, employing a moving window minimum filter to filter each color channel of the pixel, obtaining the maximum value of the color channel, and enabling the maximum value to serve as anestimation value of an atmospheric light component; calculating the chromatic aberration of the local pixels on the boundary constraint to construct a weighting function, and constructing a refined target function of the scene transmittance according to the weighting function; based on the improved guided filtering optimization objective function, outputting a final image based on the optimized transmissivity and the estimated value of the atmospheric light component; processing the final image by using contrast-limited adaptive histogram equalization, and improving local details of the localcontrast-enhanced specular highlight image. According to the method disclosed in the invention, the definition and color features of the image are effectively enhanced, and the problem that the texture information of the highlight shielded area in the image is lost is solved.
Owner:XINJIANG UNIVERSITY

Truck re-identification retrieval method based on multi-branch feature fusion

InactiveCN113609320ASolve the phenomenon of forgery, removal, and shieldingRich detail featuresCharacter and pattern recognitionNeural architecturesAlgorithmEngineering
The invention discloses a truck re-identification retrieval method based on multi-branch feature fusion, and the method specifically comprises the steps: collecting different types of truck pictures, and dividing the truck pictures into a training set and a test set; preprocessing the picture space and size of the training set, and standardizing the pictures; establishing a deep convolutional neural network structure, and training by using the loss function and the training set to obtain a network model; and according to the obtained network model, testing by using a test set, and retrieving a plurality of most similar pictures by calculating a similarity matrix of the pictures. According to the invention, the local features of the head and the compartment of the truck photo and the global features of the truck are obtained through the multi-branch feature extraction network, and then feature fusion is carried out to obtain overall feature representation. Therefore, the problems of license plate counterfeiting, removal, shielding and the like caused by difficulty in retrieval due to small truck difference in a real scene are solved. The accuracy of truck retrieval is improved. The situation that similar vehicles of the same series are difficult to distinguish is avoided.
Owner:合肥市正茂科技有限公司

A target and background fusion simulation method based on actually measured infrared imaging data

The invention discloses a target and background fusion simulation method based on actually measured infrared imaging data. The target and background fusion simulation method is mainly used for generating infrared images of a target at different backgrounds and different positions in the backgrounds. The method comprises the following steps of implementing the method; Shooting infrared images, calculating the actual size of a target in the target infrared image; calculating the radiation brightness of the infrared features generated by the target in the target infrared image, separating the target infrared image in the infrared image, calculating the infrared features of the analog target in the analog infrared imaging system, fusing the infrared features of the analog target and the background, and outputting an infrared simulation image in which the analog target and the background are fused. The method can obtain the backgrounds of the target under different conditions and the infrared images at different positions in the backgrounds, and has the advantages of high authenticity and rich detail features of the infrared target image generated by simulation.
Owner:XIDIAN UNIV

Method for generating traffic sign recognition model

According to the traffic sign recognition method provided by the invention, the detection effect on the road traffic sign is remarkably improved, and the requirement on real-time performance is met. A weighted bidirectional feature pyramid network with more balanced accuracy and efficiency is adopted to replace a path aggregation network, and channel features of road traffic signs are better fused. Secondly, common convolution is replaced by the cavity convolution, and the cavity convolution is combined with the space pooling pyramid, so that the receptive field is further expanded. Meanwhile, the detection scale is increased to four types, and the small target detection effect is improved; and random cutting is added in a data enhancement technology, so that the model learns more detail features. Finally, digital image operation technique is used to increase the number of instances for low precision categories.
Owner:SHANGHAI INST OF TECH

Pedestrian attribute identification method and identification system based on sequence context relation learning

The invention discloses a pedestrian attribute identification method and identification system based on sequence context relation learning. The pedestrian attribute identification method comprises thefollowing steps: establishing and training a pedestrian identification system; encoding the image to be identified into an image sequence in the vertical direction, and initializing an attribute sequence into a random value; calculating an image context relation sequence and an attribute context relation sequence by using the trained pedestrian recognition system; calculating the attention of theimage context relation sequence to each element in the attribute context relation sequence; and calculating the probability that each attribute belongs to each category of the attribute, and selecting the category with the maximum probability value as the category of the attribute. According to the method, the contextual relationship between the image sequences, the contextual relationship between the attributes and the contextual relationship between the images and the attributes are fully utilized, and the accuracy of pedestrian attribute recognition is improved.
Owner:HEFEI UNIV OF TECH

Fetal electrocardiogram extraction system based on convolutional encoding-decoding neural network and method

The invention discloses a fetal electrocardiogram extraction system based on convolutional encoding-decoding neural network and a method thereof. The system comprises a data collection device, a maternal electrocardiograph component estimation device, and a fetal electrocardiograph component extraction device. The data collection device is for collecting a real abdominal electrical signal of a pregnant woman. The maternal electrocardiograph component estimation device is for using a convolutional encoding-decoding neural network to estimate a maternal electrocardiograph component in the abdominal electrical signal of the pregnant woman; in training, an analog abdominal electrical signal is input into a neural network and a maternal electrocardiograph component in the analog abdominal electrical signal serves as a network label; and in testing, the real abdominal electrical signal of the pregnant woman is input into the neural network and the estimated maternal electrocardiograph component in the abdominal electrical signal is output by the neural network. The fetal electrocardiograph component extraction device is for subtracting the obtained maternal electrocardiograph component from the collected abdominal electrical signal of the pregnant woman to extract a fetal electrocardiograph component from the collected abdominal electrical signal of the pregnant woman. The method comprises the steps of data pre-processing, estimation of the maternal electrocardiograph component and extraction of the fetal electrocardiograph component. Through the technical scheme, the efficiencyand accuracy of fetal electrocardiograph extraction can be effectively improved.
Owner:SUN YAT SEN UNIV

A rapid reconstruction algorithm for a PET image

The invention discloses a rapid reconstruction algorithm for a PET image, and the algorithm comprises the following steps: carrying out the one-time iteration of a response curve which accords with atime window through an MLEM algorithm, and obtaining an image which approaches a convergence value; Performing format conversion on the generated image to obtain an RGB image; Classifying the image subjected to format conversion through a deep convolutional neural network to obtain a template of an input image; Carrying out image detail enhancement on the image by utilizing a region growing method; Carrying out point-by-point multiplication on the template subjected to image classification and the template subjected to detail enhancement to obtain a new template; Carrying out point-by-point multiplication on the image obtained after the primary iteration and a new template to obtain an input image of the next iteration; And carrying out second iteration on the image through an MLEM algorithm to complete image reconstruction. According to the method, the problems of low reconstruction efficiency, fuzzy image details, degradation and the like in positron emission tomography image reconstruction are solved.
Owner:NANJING UNIV OF AERONAUTICS & ASTRONAUTICS

Integrated phase control array fingerprint sensor

The invention discloses an integrated phase control array fingerprint sensor. The sensor includes a circuit board provided with a microcontroller, a phase control array transceiving module used for fingerprint recognition, a gyroscope, an acceleration sensor, an analog matrix switch used for signal switching, an A / D conversion module and a power source module used for power supply. The phase control array transceiving module, the gyroscope and the acceleration sensor are connected with the input terminal of the A / D conversion module through the analog matrix switch. The output terminal of the A / D conversion module is electrically connected with the microcontroller. According to the invention, phase control array principle is applied to fingerprint recognition, so that recognition rate and accuracy are improved. The gyroscope and the acceleration sensor are integrated to the phase control array fingerprint sensor in an integral design, so that the structure is simple and the size is small and influence on sensor signals by the external environment is reduced and the stability of the sensor is improved. The invention solves problems of high power consumption and incapability of waking and mobile alarm of a single fingerprint sensor.
Owner:XINFOO SENSOR TECH CO LTD

House type structure analysis method and device

The invention discloses a house type structure analysis method and device. The method comprises the steps: S1, obtaining three-dimensional object point cloud data and three-dimensional air point cloud data; S2, identifying the three-dimensional object point cloud to obtain a furniture point cloud, performing two-dimensional object projection on three-dimensional object point cloud data to obtain a rotation correction matrix, and performing rotation correction on the three-dimensional object point cloud, the three-dimensional air point cloud and the furniture point cloud according to the rotation correction matrix; S3, generating an anti-interference two-dimensional overlook angle object projection and a normal vector projection by combining the corrected three-dimensional object point cloud and furniture information to which each point belongs, and respectively generating a two-dimensional room mask, two-dimensional angular points to which each room belongs, and relevance between the angular points; S4, generating local boundaries of the rooms and global boundaries of the floors according to the angular points, and performing optimization based on the angular points and the parallel distances of the boundaries; and S5, integrating furniture position and channel position information to generate a two-dimensional house type structure.
Owner:上海诚明融鑫科技有限公司

Point cloud processing method and system, mobile platform and storage medium

The invention provides a point Cloud Processing Method, system, a removable platform, and a storage medium, by obtaining a relative positional relationship between adjacent point clouds on the timingsequence, performing aggregation processing on adjacent point clouds in the timing sequence., so as to obtain a plurality of large-point clouds, namely a plurality of aggregation point clouds, even ifthe area adjacent to the position is shielded by the shielding object, the matching calculation between the large point clouds can be completed through the detail features in the large point cloud, so that the target relative position relation between the point clouds obtained by detecting the detection equipment at different moments can be determined according to the relative position relation between the large point clouds adjacent to the positions, so that the accuracy of the relative position relation between the point clouds at different moments can be improved.
Owner:SZ DJI TECH CO LTD

Electroencephalogram signal classification method based on model uncertainty learning

The invention discloses an electroencephalogram signal classification method based on model uncertainty learning, and the method comprises the steps: 1, carrying out the preprocessing of original electroencephalogram data, including data selection, sliding window slicing, data upsampling at the early stage of attack, and data input shape selection; 2, establishing a deep learning model of the Pelee network; 3, embedding a dropout layer at the tail part of the model; 4, in a training stage, inputting data and continuously optimizing model parameters through cross entropy loss to obtain a final classification model for classification of electroencephalogram signals to be tested; and 5, in a test stage, in combination with an improved Monte Carlo discarding sampling technology based on continuous sample time information aggregation, performing prediction classification on electroencephalogram samples to be classified. According to the method, the model uncertainty is combined into a lightweight network (PeleeNet), so that the classification accuracy of the electroencephalogram signals can be remarkably improved, and the application value of the electroencephalogram signals in the fields of medical treatment and the like is increased.
Owner:HEFEI UNIV OF TECH

Rapid and effective mirror surface highlight image enhancement method and device and storage medium

InactiveCN111833271AEnhance Specular Highlight ImageRestore local informationImage enhancementComputer graphics (images)Radiology
The invention discloses a rapid and effective mirror surface highlight image enhancement method and device and a storage medium, and the method comprises the steps of processing an inputted mirror surface highlight image through employing an improved guide filtering algorithm, so as to enhance the contrast of the image, and dividing the processed image into a base layer and a detail layer; obtaining an improved cost function, namely introducing local mean square errors of all pixels into the base layer for keeping the edge of the image; extracting a brightness component in the LAB space, and enhancing the brightness component by using a contrast-limited adaptive histogram equalization algorithm; and introducing an adaptive amplification factor into the detail layer to suppress noise whileenhancing details. The image processed by the method can well recover the shielded local information in the specular highlight image, and can reserve more detail features compared with the original image.
Owner:XINJIANG UNIVERSITY

A distributed optimization method for large-scale grids

ActiveCN109147032ARun fastWork around the limitation of not being able to load all drone imagery setsImage analysis3D modellingAerial photographyComputer science
The invention discloses a distributed optimization large-scale mesh method, which distributively optimizes the initial mesh after mesh reconstruction in three-dimensional reconstruction. The method comprises the following steps: (1) dividing the input initial mesh into a plurality of mesh blocks and sending the mesh blocks to each machine of the cluster; (2) adding boundary constraints on each machine of that clust to distributively simplify and refine the mesh blocks obtained in the step 1); (3) sending the mesh blocks processed in the step (2) on each machine to the master of the cluster, fusing the mesh blocks, and simplifying and refining the mesh blocks to obtain a uniform mesh with no redundant vertices; (4) using multi-layer K-path segmentation algorithm, the aerial images of UAV being segmented into K sub-images; (5) utilizing the image set obtained in the step (4) to distributively restore the detail features of the mesh obtained in the step (3). The invention can generate a smooth mesh with rich detail features from the input initial mesh.
Owner:BEIHANG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products