88 results about "Exposure fusion" patented technology

An apparatus and method for generating a High Dynamic Range (HDR) image from which a ghost blur is removed based on a multi-exposure fusion. The apparatus may include an HDR weight map calculation unit to calculate an HDR weight map for multiple exposure frames that are received, a ghost probability calculation unit to calculate a ghost probability for each image by verifying a ghost blur for the multiple exposure frames, an HDR weight map updating unit to update the calculated HDR weight map based on the calculated ghost probability, and a multi-scale blending unit to generate an HDR image by reflecting the updated HDR weight map to the multiple exposure frames.

A method for processing high dynamic range (HDR) images by selecting preferred tone mapping operators and gamut mapping algorithms based on scene classification. Scenes are classified into indoor scenes, outdoor scenes, and scenes with people, and tone mapping operators and gamut mapping algorithms are selected on that basis. Prior to scene classification, the multiple images taken at various exposure values are fused into a low dynamic range (LDR) image using an exposure fusing algorithm, and scene classification is performed using the fused LDR image. Then, the HDR image generated from the multiple images are tone mapped into a LDR image using the selected tone mapping operator and then gamut mapped to the color space of the output device such as printer.

The invention discloses a splicing device for a real-time high speed high definition panoramic video and a method thereof. The method comprises the steps: S10, obtaining spliced multipath real-time video sequences; S20, judging whether video images at the same moment in any two paths of video sequences meet the image splicing condition; S30, calculating an image splicing parameter; S40, selecting a projection face and mapping the video images in the multipath real-time video sequences to the same mapping face according to the splicing parameter, then determining the panoramic projection face; S50, establishing a mapping table of pixel points on the video images and coordinate points of the overall projection face; S60, conducting exposure and fusion processing on pixel values of all the coordinate points on the panoramic projection face and outputting the panoramic video images according to the pixel values. The splicing device for the real-time high speed high definition panoramic video and the method of the splicing device can greatly improve splicing efficiency of the images, ensure output quality of the images, and meet the requirement for real-time high speed high definition panoramic video.

The invention provides a multi-lens panoramic stitching control method and a multi-lens panoramic stitching control device. The panoramic stitching control method comprises the following steps: firstly, a multi-channel video of a panoramic camera is read, focal length and an included angle of each lens are regulated by a motor according to vision scope and monitoring distance to be monitored to ensure that a part of superposed area is arranged between adjacent pictures. And then, after the focal length is adjusted, the included angle between lenses is correspondingly regulated to carry out space matching alignment onto a superposed video sequence, and effect of forming a large-viewing-angle high-solution continuous image containing video sequence information is realized through exposure fusion.

The invention provides a method for synthesizing a ghost-reduced high-dynamic-range (HDR) video based on a block matching dynamic estimation algorithm. The method comprises the following steps: block matching: selecting a reference frame and a current frame from an input picture sequence, performing gray-scale processing, and calculating a motion vector of the current frame through an ARPS3 block matching algorithm; detection and calibration of motion areas: detecting respective motion areas of the current frame and the reference frame through the calculated motion vector, combining the motion areas of specific images, and marking on a binary bitmap; and synthesis of a ghost-reduced HDR image: improving a weight map obtained by calculation according to an Exposure Fusion algorithm by the obtained binary bitmap, and fusing images needing to be fused under the guidance of the improved weight map in order to obtain a final HDR image, namely, video frames. According to the HDR video synthesis method, ghost areas are detected through motion estimation, and ghosts are reduced through a fusion algorithm to synthesize the HDR video frames.

The invention provides a pointer instrument reading automatic identification method based on a single-point multi-box detector. The method comprises the following steps of Generating an enhanced imageby adopting a low-light enhancement method based on a multi-exposure fusion framework, and recovering a clear image from the blurred image by adopting a two-stage kernel estimation based motion blurremoval method; Designing a dial detector SSD structure based on VGG-16 network and convolutional neural network to detect the dial in the image, output dial position and confidence information ; Constructing a training data set, training SSD network parameters by using the training data set, and setting related parameters and a learning strategy; Accelerated robust features of the image are extracted to generate feature point descriptors, and the rotating dial plate image is corrected according to the feature matching technology; Detecting and extracting an edge image by using a CannyLines line segment detection algorithm based on a gradient amplitude, collecting a collinear point group from the edge image and fitting the collinear point group into a straight line final verification linesegment; And determining the optimal line segment as the real position of the pointer, and calculating the reading of the pointer in the dial plate by using the relationship among the angles.

The invention provides a high dynamic range imaging method for removing ghosts through moving object detection and extension. The important thought of the method of the invention is to extend a moving area under edge constraints to enhance a moving area detection result. A Markov random field-based method is used for detecting the moving area, an image segmentation-based method is used for extending the moving area, the weight of each pixel is adjusted according to a mask image of the moving area after extended, the weight of each pixel is applied to final exposure fusion, and the problem of ghosts appearing in the high dynamic range imaging can be effectively solved.

The invention discloses a single image enhancement method based on a full convolution neural network. Firstly by considering that the lost information of a single exposure image can be described by amulti-exposure sequence, a low-exposure image and a high-exposure image different from the input image exposure are generated by using the constructed predictive exposure network, so as to construct amulti-exposure sequence. Secondly, in order to avoid the problem of low robustness caused by manual feature extraction, the exposure fusion network is used to complete feature extraction, feature fusion and image reconstruction. Finally, by considering that the deconvolution layer in the predictive exposure network will cause the chessboard artifacts, the structure anisotropy loss related to human perception is used to train the predictive exposure network to alleviate the artifacts. The invention has the advantages that the whole contrast of the image can be improved, and certain informationof the underexposed and over-exposed regions of the image can be recovered.

The invention relates to the field of computer image high-dynamic-range imaging technology, and provides an exposure fusion method in consideration of brightness distribution and detail presentation. The method comprises the steps of decomposing a low-dynamic-range multiple-exposure-image sequence into a Lab color space, and obtaining a brightness sequence and a color sequence; calculating a brightness global weight function according to the brightness sequence; segmenting the brightness sequence on a line number and row number plane for obtaining a plurality of sub-pixel calculating units, and obtaining a brightness detail weight function, segmenting the color sequence on the line number and row number plane for obtaining a plurality of sub-pixel calculating units, and obtaining a color detail weight function; performing weighted summation on the brightness sequence by means of the brightness global weight function and the brightness detail weight function; and performing weighted summation on the color sequence by means of the color detail weight function for obtaining a fused color component. The exposure fusion method realizes single high-dynamic-range quick imaging from a plurality of low-dynamic-range images and furthermore satisfies a requirement for brightness distribution and detail presentation.

An apparatus and method for generating a High Dynamic Range (HDR) image from which a ghost blur is removed based on a multi-exposure fusion. The apparatus may include an HDR weight map calculation unit to calculate an HDR weight map for multiple exposure frames that are received, a ghost probability calculation unit to calculate a ghost probability for each image by verifying a ghost blur for the multiple exposure frames, an HDR weight map updating unit to update the calculated HDR weight map based on the calculated ghost probability, and a multi-scale blending unit to generate an HDR image by reflecting the updated HDR weight map to the multiple exposure frames.

A method and an apparatus for image filtering are described. Structural information is employed during the calculation of filtering coefficients. The structural information is described by the regions defined through an edge map of the image. In one embodiment, the region correlation between the target pixel and a contributing pixel is selected as a structural filtering coefficient. The region correlation, which indicates the possibility of two pixels being in the same regions, is calculated by evaluating the strongest edge cut by a path between the target pixel and a contributing pixel. The structural filtering coefficient is further combined with spatial information and intensity information to form a spatial-intensity-region (SIR) filter. The structural information based filter is applicable to applications such as denoising, tone mapping, and exposure fusion.

The invention discloses a direct multi-exposure fusion parallel acceleration method based on the OpenCL; the method comprises the following steps: using a CPU+GPU heterogeneous parallel framework and the GPU strong floating point calculating capability, fusing cores under the OpenCL exploitation environment, caching a Gaussian kernel to a constant memory, caching reuse data to a local memory, and increasing the work load of each work item, thus shortening an access and calculation time, and realizing faster multi-exposure fusion. Compared with a conventional serial processing method, the method can obtain the maximum speed-up ratio reaching 11.19, thus effectively reducing the multi-exposure fusion algorithm operation time, and providing strong guarantee for further applications of the multi-exposure fusion algorithm; in addition, the method can obviously improve the space frequency and average gradient, thus effectively solving the scene detail information loss problems caused by insufficient electrography and display equipment dynamic scope.

The invention discloses a multi-exposure image fusion method of automatically removing ghosts. The multi-exposure image fusion method of automatically removing ghosts provides an effective method of moving object selection, for the phenomenon that ghosts exist in multi-exposure image fusion of a dynamic scene. The multi-exposure image fusion method of automatically removing ghosts includes the steps: selecting one input LDR image as a reference image, detecting the motion area by means of a method of bidirectional luminance mapping, calculating the motion weight, and automatically removing theghosts by means of a method of gradient domain weighted optimization so as to obtain a latent image; and after removing the ghosts, calculating the fusion weight by means of the fusion method based on blocks, performing exposure fusion, maintaining the detailed information and suppressing the abnormal value. The multi-exposure image fusion method of automatically removing ghosts can effectively maintain the details of a plurality of input images and remove the ghosts, can avoid complicated parameter setting and arbitrariness for a user to set the threshold, and has higher robustness for the ghost phenomenon in various scenes.

The invention discloses a single-frame high-dynamic imaging method based on a convolutional neural network. The method comprises the following steps: constructing an enhancement module; constructing an initial convolutional neural network with an enhancement module according to the enhancement module; training the initial convolutional neural network to obtain a trained convolutional neural network; and processing the original single-frame low-dynamic image through the trained convolutional neural network to generate a high-dynamic image. According to the method, the high-dynamic image can bedirectly generated from the single-frame low-dynamic image, the problem that multiple frames of exposure images cannot be obtained at the same time by a moving video in multi-exposure fusion is solved, parameter adjustment is not needed, and the method is convenient to use.

The invention discloses a saturation based quality evaluation method for a colored multi-exposure fusion image, and the method can be used to solve the problem that evaluation for the colored multi-exposure fusion image is inaccurate. A multi-exposure image in an MEF database and a fusion image of the multi-exposure image serve as a training sample, and texture, structure and color information is obtained from the training sample by utilizing a saturation and wavelet coefficient based extraction manner; texture, structure and color similarities are calculated respectively; all the obtained similarities, as characteristic values, together a given MOS value are input to an ELM machine learning device to implement training; and the texture, structure and color similarities of the multi-exposure fusion image and the corresponding generated fused image are input to the trained ELM machine learning device, and an evaluation result is obtained.

The invention relates to an image processing method and system based on the multi-exposure fusion and the backtracking algorithm. The image processing method based on the multi-exposure fusion and the backtracking algorithm comprises the following steps: obtaining the original image data in the high dynamic photographing mode; performing coarse positioning on the original image data by means of the backtracking algorithm and obtaining a key frame; performing accurate positioning on a number of image frames starting from the key frame by means of the Euclidean distance, and obtaining the optimal image data; judging whether the optimal image data is obtained completely or not; if not, returning to the step of performing coarse positioning on the original image data by means of the backtracking algorithm and obtaining the key frame; if so, performing image fusion on the optimal image data by means of the multi-exposure fusion algorithm, and obtaining the fusion image data combined with the Laplasse product processing. By means of the image processing method and system based on the multi-exposure fusion and the backtracking algorithm, the dynamic folding and the artifact problem caused by the arm jitter is effectively restrained, the edge blur problem is eliminated, the processing speed of the data is improved, the real-time performance of the HDR photographing is ensured, and the appearance of black screen and dead machine under the pressure test of the camera is avoided.

A method, device and storage medium for enhancing image contrast, the method comprising: invoking a neural network, the training set of the neural network being a group image pair, wherein each pair of images includes a first image and a second image for the same scene, a contrast of the first image is lower than a contrast of the second image (101); a third image is input to the neural network (102); and a fourth image outputted through the neural network map is obtained, the fourth image The contrast is higher than the contrast of the third image (103). The method can enhance the low-contrast image of any input, so that it can achieve the high dynamic range of the multi-exposure fused image, so the contrast-enhanced image is real and the image quality is high.

The invention provides a double exposure fusion method and device for an image. The double exposure fusion method for the image comprises the steps of performing human face control point recognition on an original face image, and obtaining a face region graph; performing brightness filling on the face region graph, and obtaining a face weight graph; performing weight fusion on the face weight graph and a single-channel image of the original face image, and obtaining a fusion weight graph; performing soft light processing on the original face image and an original background image, obtaining a soft light processing graph, performing color filter processing on the original face image and the original background image, and obtaining a color filter processing graph; performing fusion processing on the soft light processing graph and the color filter processing graph, and obtaining an image to be fused; and performing fusion processing on the original face image and the image to be fused according to the fusion weight graph, and obtaining a double exposure fusion image. According to the double exposure fusion method provided by the invention, the double exposure effect of the face image in the double exposure fusion image is better through the above manner, so that the double exposure fusion image is more nature.

The invention provides an HDR image generating method based on self-adaptive double gamma transform. The method comprises the steps of: firstly, based on the acquisition of the double gamma transformvalue of the statistical property of images, calculating the double gamma transform value of an input LDR original image; secondly, performing gamma transform on the input image; thirdly, performing exposure fusion on the two generated images. Compared with the prior art, the method can rapidly re-lock a current converter under the condition of small circuit currents, thereby effectively preventing reclosure from causing secondary overcurrent impact on a system and achieving a positive effect of low current harm. The invention provides a simplified algorithm for converting an input LDR image to an image with an HDR effect, and the implementation process is simple and the complexity is low; the method is free of manual parameter testing and can obtain corresponding transform parameters through self-adaptive calculation for different input images, so that the process is more intelligent.

The invention discloses a multi-exposure image fusion method based on Retinex theory, belonging to the technical field of image processing. The method comprises the following steps: estimating brightness of exposure sequence images according to Retinex theory to generate brightness image Lk and reflectance image Rk of each image; The final brightness image is obtained from the plurality of brightness images Lk and the defined S curve shown in the description; The reflectance image Rk was evaluated to obtain the final reflectance image shown in the description. Then by shown in the descriptionand shown in the description, and then the contrast of the fused image is enhanced by using the automatic color scale method, and the final fused image is obtained. The invention applies Retinex theory to multi-exposure image fusion of static scene, Trying to get the real scene brightness image and reflectivity image from the exposure sequence image can effectively solve the problems of detail loss and color distortion in the traditional multi-exposure fusion method, so that the overall brightness of the final fusion image is more real and the visual effect is better.

The invention discloses a double-channel exposure fusion network model and method for low-illumination image enhancement. The network model comprises a first enhancement branch, a second enhancement branch and a fusion module. Wherein the first enhancement branch and the second enhancement branch are used for processing a to-be-enhanced low-illumination image by using different preset enhancementstrategies respectively; and the fusion module comprises an attention unit and a fine adjustment unit and is used for carrying out weighted fusion on the enhancement results obtained by the first enhancement branch and the second enhancement branch to obtain an enhanced image. In addition, the invention also provides a denoising strategy divided into two steps, so as to carry out efficient adaptive suppression on the noise in the enhancement process. The model provided by the invention has clear physical explanation and is light and effective. The network model can be used for low-illuminationenhancement tasks in various low-light environments, especially in consideration of changes of imaging environments and high requirements for processing real-time performance.

The invention relates to an edge detection and frame difference method-based high-dynamic range video de-ghosting method. Image registration-based edge detection is performed on interlaced images with the same exposure rate; after image registration, higher accuracy can be obtained through using a motion detection method; computational complexity is reduced based on a motion detection-based frame difference method; adjustment is carried out at some extreme boundary positions through using a weight map, so that bad special values can be eliminated; and when a traditional fusion exposure method is adopted, the weights of some boundary positions may be 0, and the weights of positions adjacent to the boundary positions are not 0, and this difference may make a final high-dynamic range video be low in accuracy at some positions, while, according to the method of the invention, offset is added, so that weight values will be more smooth.

The embodiment of the invention relates to the field of image processing, and discloses a shooting method, system and device, electronic equipment and a storage medium. In a part of embodiments of theinvention, the shooting method comprises the following steps: determining N areas needing to be exposed in a scene; Respectively exposing each area, and when each area is exposed, shooting a pictureof at least one scene; And carrying out exposure fusion on the shot photo to obtain a final scene photo. In the implementation process, photos of which each area needing to be exposed is exposed are shot, the shot photos are subjected to exposure fusion, and a final photo is obtained, so that the final photo is closer to an image of the effect of a real scene seen by human eyes.

The invention provides a multi-exposure image fusion method based on an attention generative adversarial network. The thought of the attention mechanism is highly matched with the detail weighting problem in multi-exposure fusion, the weight of each input image can be adaptively selected by applying channel attention, and the weights of different spatial positions are adaptively selected by usingspatial attention. The technology has a wide application prospect in various multimedia vision fields. According to the algorithm, a new attention generative adversarial network is designed to be usedfor a multi-exposure image fusion task, and a visual attention mechanism is introduced into the generative network, so that the network can be helped to adaptively learn weights of different input images and different spatial positions to achieve a better fusion effect.

Provided is an image fusion method, device and unmanned aerial vehicle. The method comprises following steps: performing bit plane layering on an original image of a frame to obtain a plurality of bitplanes of the original image, wherein the bit number of the original image is M, and M is an integer greater than 1; obtaining a plurality of frames of reconstructed images according to a plurality of bit planes, wherein the bit planes corresponding to different reconstructed images are not identical, and the bit number of each reconstructed image is N, and N is a positive integer smaller than M;fusing multi-frame reconstructed images to obtain a fused image of the original image. As the multi-frame reconstructed images directly come from the same frame image, thus ensuring the consistency of the image scene, it can be directly applied to exposure fusion without image registration, and the resulting fusion image will not contain ghost images. Compared with the prior art, the invention can avoid image blurring conveniently, highlight darkness and highlighted details at the same time, on the other hand, can reduce the algorithm resource overhead of image fusion.

A method and an apparatus for image filtering are described. Structural information is employed during the calculation of filtering coefficients. The structural information is described by the regions defined through an edge map of the image. In one embodiment, the region correlation between the target pixel and a contributing pixel is selected as a structural filtering coefficient. The region correlation, which indicates the possibility of two pixels being in the same regions, is calculated by evaluating the strongest edge cut by a path between the target pixel and a contributing pixel. The structural filtering coefficient is further combined with spatial information and intensity information to form a spatial-intensity-region (SIR) filter. The structural information based filter is applicable to applications such as denoising, tone mapping, and exposure fusion.

The invention discloses a multi-exposure image fusion method based on depth perception enhancement. Firstly a bidirectional contrast enhancement strategy of an image is provided, the enhancement processing of each source image is provided, and a series of detail-enhanced reference images are obtained; and a detail enhancement network DEM is constructed, a loss function is designed, each reference image is weighted according to information richness to serve as an optimization target, and the detail enhancement network is trained, so that fusion brightness containing rich details is obtained. A color enhancement network CEM is constructed, and a loss function is designed, so that the color enhancement network CEM learns a mapping relationship of colors with respect to brightness in different scenes, thereby realizing color rendering for brightness fusion. The two networks respectively generate brightness and color components to jointly complete generation of a fused image. The multi-exposure fusion image obtained by using the method not only has rich detail information and good global structure consistency, but also has more real and vivid color information compared with the existing method, and the visual quality of the image is remarkably improved.

The invention discloses a non-uniform illumination image enhancement method based on virtual multi-exposure fusion, and belongs to the field of image processing. The method comprises the following steps: 1) converting an input image from an RGB space to an HSV space; 2) for the V component, determining two optimal exposure ratios based on an image information entropy maximization principle, and performing two times of virtual exposure by using a brightness mapping function to generate an enhanced exposure image V1 and a weakened exposure image V2; 3) fusing an exposure image sequence consisting of V, V1 and V2 by adopting a multi-exposure image fusion method with detail lifting to obtain an enhanced V component; and 4) re-synthesizing an HSV image by using the enhanced V component and the unchanged H component and S component, and converting the HSV image back to the RGB space to serve as a final result to be output. According to the invention, the image quality problem caused by overexposure and underexposure areas existing in the non-uniform illumination image at the same time can be effectively improved, and the image with a high dynamic range and capable of well maintaining color and detail information is generated.

The invention relates to a full reference multi exposure fusion image quality evaluation method based on contrast and saturation. The full reference multi exposure fusion image quality evaluation method based on contrast and saturation comprises the following steps: fusing a plurality of reference images as a new reference image using the contrast as the weight based on the multi-scale decomposition by fully considering the characteristics of the human visual system and the multi exposure images and utilizing the phenomenon that human eyes are sensitive to the contrast and saturation of the fused image; performing similarity comparison on the contrast and saturation of the multi exposure fusion image to be evaluated to obtain the single layer quality evaluation value; weighting and fusing the multiple layers of quality evaluation values to obtain the final score. The full reference multi exposure fusion image quality evaluation method based on contrast and saturation is simple and fully reflects the subjective perception of the human eyes to the multi exposure fused image, and can better evaluate the quality of the multi exposure fused images.

The invention discloses a low-light-level image enhancement method based on an MSR theory and exposure fusion, and solves the problems of damage to the overall structure of an image, excessive enhancement and the like in low-light-level image enhancement. The method comprises the steps of inputting an original image and extracting a V component of the original image; processing the V component byusing a local detail processing operator to obtain a detail processing graph; processing the V component by using an integral structure preserving operator to obtain a structure processing graph; defining the V component of the original image as a correction processing image; combining the three processing images by adopting an exposure fusion method to obtain an enhanced result image; and transferring from the HSV space to the RGB space for display to obtain an enhanced result graph. According to the invention, the regional detail richness based on the information entropy is used for accurately measuring the detail richness of the image, three different Gaussian kernel weights are adaptively calculated according to the regional detail richness, and three processing images are merged by combining an exposure fusion method. According to the method, image detail enhancement and structure and brightness maintenance are considered while the low-light-level image is enhanced. The method canbe applied to the fields of remote sensing, military, industry and medicine.