118 results about "Non region of interest" patented technology

The invention relates to the field of a video coding technology, and discloses a region of interest (ROI) video coding method and apparatus based on object detection. The video coding method comprises the following steps: performing the object detection on an original image, generating object information, and obtaining an object area and confidence; converting the object area into an ROI suitable for coding; according to coding quality grades and the confidence, calculating coding quantification parameters; and according to the corresponding coding quantification parameters, performing quantification coding on the ROI and a non-ROI. According to the invention, the ROI, the non-ROI and the corresponding confidence are obtained through the object detection, and the corresponding quantification parameters are selected for performing the quantification coding on the ROI and the non-ROI, such that the coding quality and stability of the ROI can be effectively ensured, at the same time under the condition that the coding quality of the ROI is not reduced, the compression code stream of a whole image can be decreased, and the video transmission and storage efficiency is improved.

The invention provides an HEVC (High Efficiency Video Coding) code rate control method based on region-of-interest. The HEVC code rate control method based on region-of-interest comprises the steps: generating a spatial domain significance graph of the current frame according to a GBVS (Graph Based Visual Saliency) model; generating a time domain significance graph of the current frame through motion vector information; integrating the time domain significance graph and the spatial domain significance graph by using the consistency normalization method to obtain the eventual significance graph; performing regional division for the current frame image by using the significance graph so as to divide the region into a region-of-interest and a non-region-of-interest; performing bit distribution for the region-of-interest and the non-region-of-interest respectively; performing bit distribution for each LCU in the current frame according to the significance; calculating lambda and a QP (Quantization Parameter) value according to the distributed code rate, and clipping and correcting the lambda and the QP value; and performing coding by using the eventually obtained lambda and the QP value. The HEVC code rate control method based on region-of-interest can improve the subjective quality for video coding and accurately control the output bit at the same time.

The invention relates to spatial domain resolution adjustable coding method based on interested domain and corresponding decoding method, the coding method is that loading information of the interested domain, dividing the original image into an interested domain and a un-interested domain according to information of the interested domain, coding image in the interested domain according to resolution of the original image, coding image in the un-interested domain by resolution lower than that of the original image; the decoding method is that obtaining information of the interested domain from the coding code stream, dividing the macroblock code stream to be decoded in to interested domain decoding content and un-interested domain decoding content, and decoding the interested domain decoding content according to standard of high-definition images. The invention realizes spatial domain resolution adjustable coding method based on interested domain and ensures quality of the interested domain video information while greatly reduces number of bits for transmission of coded images, which is especially suitable for safety protection monitored transmission requirements.

The invention discloses a method, a device and a system for regional division / coding of an image. By extracting a motion vector of the image in a video sequence, judging the image as an interest region or a non-interest region according to the complexity of the motion vector, and performing further accurate judgment on the image after regional division according to spatial correlation and temporal correlation of a macro block, the accurate division of a foreground region and a background region is performed on the image. A smaller quantization parameter is adopted for the interest region to improve the quality of a video, and a higher quantization parameter is adopted for the non-interest region to balance the overall bit consumption to be unchanged, and finally achieve the effect of improving the subjective quality of the video.

The invention is suitable for the technical field of intelligent traffic and provides a method and a system for detecting lane marked lines. The method comprises the following steps of: obtaining an input original image and performing binaryzation on gray level of the input original image; utilizing a canny algorithm to obtain an edge image of the image after being subjected to gray level binaryzation; removing non-interesting area in the edge image; scanning each pixel point of the image after being processed; if the point is an edge point, storing the coordinate of the point; taking a straight line for each degree, thereby obtaining initial left and right lane marked lines; taking the straight line meeting a first preset condition and having the most edge points according to the initialleft and right lane marked lines; taking the left and right lane marked lines meeting a second preset condition; sequencing the taken left and right lane marked lines; screening out the final left and right lane marked lines according to a line_X difference of the left and right lane marked lines; and taking the sobe1 edge points in adjacent areas of the screened left and right lane marked lines as the final accurate lane marked line points. According to the method provided by the invention, the speed and the precision of detecting the lane marked lines can be efficiently increased.

Techniques pertaining to scalable video codec are disclosed. According to one aspect of the present invention, a video image is analyzed and a region of interest (ROI) and a region of non-interest (non-ROI) are identified. By comparing the non-ROI image with that of a previous image, a background ignored identifier is created indicating whether the non-ROI can be ignored during encoding and decoding processes. Based on the status of the background ignored identifier, the encoder encodes the images into a basic layer (BL) and an enhanced layer (EL), and transmits the coded bit streams along with the identifier to a decoder. The decoder reconstructs the image based on the identifier and the BL and the EL bit streams.

The invention publishes a medical image ROI compression method based on a lifting wavelet and a PCNN, which comprises the following steps of: circling a region of interest by a doctor, and separating the region of interest from a region of no interest by a difference image method; adopting lossless compression in the region of interest, constructing compactly supported biorthogonal wavelet transformation through a lifting scheme, and then carrying out Huffman encoding; adopting lossy compression in the region of no interest, segmenting gray-value pixel approximate points through the PCNN, carrying out ignition operation, and then carrying out run-length encoding; and finally carrying out inverse transformation restoration, merging the region of interest and the region of no interest, and eliminating a boundary discontinuity problem through linear interpolation. Experimental results prove that the region of interest can be flexibly selected and controlled by the compression method, the used information for doctor diagnosis can be completely reserved, and the compression ratio is higher. Meanwhile, the computation for an ROI mask and the computation and the encoding for a wavelet coefficient difference value are omitted, the compression and decompression time and the algorithm complexity are reduced, and the image processing and transmitting efficiency is improved.

The invention discloses a compression and decompression method of a panoramic image, an apparatus and a system thereof. On one hand, through the compression method of the panoramic image, a compression image is acquired. The method comprises the following steps of dividing the panoramic image into an interesting area and an uninteresting area; carrying out compression processing on the uninteresting area; and synthesizing the uncompressed interesting area and the compressed uninteresting area into the compression image. On the other hand, through the decompression method of the compression image, a decompression image is acquired. The method comprises the following steps of extracting the uncompressed interesting area and the compressed uninteresting area in the compression image; carrying out decompression processing on the uninteresting area; and synthesizing the decompressed uninteresting area and the interesting area into the decompression image.

A depth processing method, an electronic device and a medium are provided. The depth processing method includes: obtaining a color image and a depth map corresponding to the color image; extracting a plurality of regions from at least one of the depth map and the color image; obtaining region information of the regions, and classifying the regions into at least one of a region-of-interest and a non-region-of-interest according to the region information, wherein the region information comprises area information and edge information; and adjusting a plurality of depth values in the depth map according to the region information.

The invention discloses a method for detecting remote sensing image change based on interest areas, which belongs to the fields of analyzing and processing remote sensing images. The invention mainly solves the problems of pseudo-change information existing in the methods for detecting remote sensing image change, the implementation process comprises: firstly, extracting the outline of a differential chart, secondly, defining the outline map of a closed interest area based on the threshold value outline map chained method, and obtaining the interest area after expanding the outline, thirdly, obtaining a new differential chart through updating the differential chart according to the gray level, the space position and the sort feature of the interest area and a non-interest area, fourthly, dividing the interest area in the new differential chart, and obtaining a change result chart. The method has the advantages of accurate location of the interest area, noise and registration error robust, less pseudo-change information and high change detection, and can be used in the fields of disaster surveillance, land utilization and agriculture investigation.

An X-ray imaging apparatus includes an X-ray source configured to radiate X-rays to a region of a subject, an X-ray detector configured to acquire a plurality of frame images related to the region of the subject by detecting the radiated X-rays, a filter configured to filter the X-rays radiated from the X-ray source, an image processor configured to set a region of interest within the region of the subject based on the plurality of frame images, and a controller configured to control the filter so that X-rays of a lower dose than a dose of X-rays made incident on the region of interest are made incident on a region of non-interest within the region of the subject, and control the X-ray detector so that a gain of the X-ray detector in the region of non-interest is greater than that in the region of interest.

The invention relates to the field of urban three-dimensional models, and specifically relates to a graphic image fusion mass building drawing method. The method comprises the steps: firstly drawing an AOI (area of interest) through a real-time graphic method on the basis of LOD and internal and external storage scheduling; dynamically generating texture images for a secondary AOI and a non-AOI according to the position of a visual point, and carrying out the screen fusion, thereby achieving the complete rendering of a scene. The method maintains the roaming continuity of graphic rendering, and the image rendering is independent of the scene complexity. Meanwhile, the method can effectively improve the bearing capability of a system under the same hardware condition, also can achieve the scene roaming of hundreds of thousands of buildings under the hardware condition of relatively lower performances, and can achieve the complete drawing of a scene without visual loss.

The invention discloses a method for establishing an urban road vehicle image database, which mainly solves the problem that the existing vehicle database is not complete and cannot effectively describe an urban traffic condition. By adopting an implementation scheme, the method comprises the following steps: determining a data acquisition scheme and acquiring an original video of a road vehicle according to factors of vehicle types, different scenes, photographing angle, illumination variation and weather condition; converting the original video into an image, and defining an interested area in the image by utilizing a trapezoidal frame; performing fuzzy processing on the non-interested area outside the trapezoidal frame; marking a vehicle image in the interested area, and generating a txt-format tag file; and converting the txt-format tag file to a corresponding xml-format tag file to be stored, and acquiring the vehicle database. By adopting the method, information about a vehicle position and type can be accurately given, reliable data support can be provided for training a deep convolutional neural network, and the method can be used for an intelligent traffic system.

The method comprises: making the conversion for the intra-frame coding frame; only making the transformation from the compress domain digital cosine transform coefficient to the integer transform coefficient; making transformation for the compress domain intra-frame prediction. Making detection for the interested area in the compress domain: the summation of absolute values of the horizontal component and vertical component of current macro block motion vector is less than or equals the threshold TH, then the macro block is estimated as an interested area. Making transformation for the non-interested area: making compress domain open-loop transformation; by using the transformation from the intra compress domain digital cosine transform coefficient to the integer transform coefficient, making re-quantization. Making transformation for the interested area: in the compress domain, making quantization error compensation; making the interpolation error compensation close-loop transformation.

The invention relates to a method for using ROI image compression to edit large image. Wherein, said method comprises: opening the MMS edit interface of mobile phone to obtain the image inserted into MMS, to compare the size of said image, and if it is over the limit, then opening the image edit interface, to reduce the size to match the screen of mobile phone; according to the user operation, selecting the shape and size of preferred section ROI and selecting the compression qualities of preferred section ROI and non-preferred section, to process image data compression, to insert the compressed image into MMS. With said inventive method, the better effect of compressed image can be obtained than JPEG, while it can reduce the memory capacity of image, and meet the minimum demand of space limit of MMS.

Improving diagnosis performance by matching the appearances of images while maintaining difference arising from the difference in tissue structure without being influenced by a non-interest region. Two medical images are divided into a plurality of tissue regions, each with each tissue imaged therein, and a similarity level is calculated for each corresponding pair of the plurality of tissue regions between the two medical images based on the form of each tissue region. An interest level is set for each corresponding pair and a weigh is obtained for each corresponding pair according to the similarity level and interest level. An image processing condition is set such that the matching level of an image characteristic of each corresponding pair is varied according to the weight, and image processing is performed on at lest either one of the two medical images according to the set image processing condition.

The invention relates to a lung parenchyma area surface model establishment method based on a computed tomography (CT) image. The method includes the steps of conducting gaussian smoothing operation on each original image of a lung CT scanned image by using a computer so as to eliminate image noise interference, obtaining a lung parenchyma area by using a three-dimensional FastMarching algorithm, using a MarchingCubes algorithm, confirming specific morphological characteristics of lung parenchyma by using calculation of surface curvature, removing a confirmed uninterested area, and finally obtaining a complete surface model of a lung interested area with a smooth surface with marginalized repair. The invention provides a computer aided imaging diagnosis method which can eliminate diagnosis difference caused by different subjective factors such as subjective experience and observational ability of doctors, provide referential identified diagnosis results with high accuracy rate, thereby enable diagnostic imaging to be more objective and improve efficiency and accuracy of diagnosis.

The invention discloses an image recognition method. The method includes: acquiring a to-be-recognized image, and determining a region of non-interest in the to-be-recognized image; acquiring initialresolution of the region of non-interest, and carrying out resolution reducing processing on the initial resolution of the region of non-interest according to a preset parameter; and carrying out image recognition calculation on the to-be-recognized image, after reducing the resolution of the region of non-interest, to recognize a target object. The invention also discloses an image recognition device and a computer-readable storage medium. The method can decrease a calculation amount and improve recognition efficiency when image recognition is carried out through deep learning.

The invention provides a generation method of a three-dimensional image. The generation method of the three-dimensional image comprise, firstly, obtaining an original two-dimensional image to serve as a first view, extracting a plurality of object regions from the original two-dimensional image, and dividing the plurality of object regions into object regions of interest and object regions of non-interest; secondly, performing overall offset on every object region of non-interest according to depth of field of different object regions of non-interest; thirdly, performing fine offset on the regions of interest; and fourthly, generating a second view according to the offset regions of interest and regions of non-interest. According to the technical scheme of the generation method of the three-dimensional image, a three-dimensional image offset amount calculation mode aiming at different showing occasions are provided, and a good three-dimensional visual immersion effect is achieved through a mode of combining regional offset and the fine offset and one the basis of reducing the calculation; meanwhile, insertion and movement of an object of interest can be performed freely and flexibly according to requirements of users, and a more convenient and fast conversion mode for three-dimensional manufacturing of movies, televisions and animations is provided.

The invention discloses a compression method for an interest region of a depth map, relating to the compression method for the depth map. According to the compression method, the problem that automatic selection of the interest region cannot be realized by a traditional compression method for the interest region of the depth map is solved. The compression method comprises the following steps of: acquiring a gray scale edge of the depth map by using an edge detection method; acquiring an edge of the depth map and a perimeter region thereof by using mathematical morphology expanding operation; carrying out region segmentation on the depth map by using an image segmentation method; solving an average gray value of pixels in each segmented region and taking the average gray value as a gray value of the region; determining the interest region according to the edge of the depth map and the perimeter region thereof and the average gray value of the pixel; carrying out smoothing processing on a non-interest region by adopting a Gaussian smoothing filter method combining time domain with air space; and coding the edge of the depth map and the perimeter region thereof by adopting a lossless compression mode. The compression method disclosed by the invention can be widely applied to compression for automatic selection of the interest region of the depth map.

A region-of-interest (ROI) video-coding method and apparatus based on fuzzy logic control for a video encoder is provided. Providing an image having a plurality of region-of-interest regions and a plurality of non-region-of-interest regions, the first step is to separate the region-of-interest regions and the non-region-of-interest regions from the image. Then by sending an input from the region-of-interest regions to a fuzzy logic control, in which the fuzzy logic control performs fuzzy manipulations that enhances the quality of the region-of-interest regions, and thereof the overall quality of an output image. The method and apparatus are particularly useful in videophone and videoconferencing.

The invention relates to the technical field of image processing and discloses a method and a device for realizing conversion from a two-dimensional image to a three-dimensional image. The method comprises the following steps of: inputting the original two-dimensional image; determining interesting areas in the original two-dimensional image for a user; respectively using the interesting areas inthe original two-dimensional image for the user as prospect areas, using non-interesting areas as background areas and generating a prospect histogram and a background histogram; calculating color difference distribution of the prospect histogram and the background histogram; generating a depth map according to the color difference distribution; generating a target two-dimensional image accordingto the depth map; and generating a binocular three-dimensional image by fusion of the original two-dimensional image and the target two-dimensional image. According to the embodiment of the invention, the scheme implementation is simple, and the real-time requirement can be met.

The invention discloses an adaptive video coding method and a terminal. The method comprises the steps: a first terminal learns a region concerned by a user of a second terminal in a screen of the second terminal, and according to the region concerned, determines a region of interest ROI and a non-region of interest NROI in a video frame acquired by the first terminal; the first terminal performs video coding for acquired video frame data according to the ROI and the NROI to realize that a code rate of a macro block set in the ROI is greater than the code rate of the macro block set in the NROI in the coded video stream; and the first terminal performs video stream encapsulation after video coding, and further sends the encapsulated data to at least one terminal comprising a video call connection with the first terminal, wherein the at least one terminal includes the second terminal. The method provided by the invention can truly reflect a true region of interest of a user of an opposite terminal in a process of performing video call by the terminal, and provides good video call visual experience for the user.

The invention relates to a mixed digital image halftoning method based on a significance visual attention model, and belongs to the technical field of image processing. The method is characterized by adopting a bottom-up visual selective attention model, and carrying out calculation to obtain image intensity, color and direction feature patterns from an input image through Gaussian pyramids and a "center-surround" operator; carrying out normalization on the feature patterns to enable the feature patterns to be superposed into a total significance pattern, and extracting regions of interest (ROIs) of the image; carrying out image halftoning on the images in the ROIs through a model-based weighted least square halftone iteration method; carrying out halftone image conversion in non regions-of-interest (NROI) through a tone-based error diffusion method, and carrying out halftone parallel computing in the two regions of the image; and objectively evaluating digital image halftoning performance through a quality evaluation method based on the selective attention model, and analyzing algorithm complexity to obtain an optimum halftone image.

The invention discloses an automatic ROI detection method of a digital pathologic full slice image. The method is applied to a digital pathologic full slice image database, and comprises an offline training stage and an online detection stage. A doctor observes full slices to set an ROI viewport in each full slice in a standard extraction database, the ROI viewpoints are spliced to obtain a complete ROI, an overlapped slide window method is used to extract a rectangular area from the ROI, the rectangular area serves as a positive sample, a rectangular area collected from a non ROI serves as a negative sample, and the samples are used to train a classifier; and the classifier is used to determine a rectangular area extracted from unknown full slices and return a result. According to the invention, rectangular area samples can be obtained by only recording a slice observing operation of the doctor, the sample obtaining difficulty can be greatly reduced on the premise that the correctness is ensured, automatic detection for full slice ROI is easy to realize, time and labor are saved, and detection is rapid and accurate.

The invention provides a visual saliency based perception rate control method and relates to the technical field of a video. According to the method, a saliency value of each pixel is calculated; thesaliency value of an LCU (Largest Coding Unit) is calculated; the saliency value of the LCU is taken as distortion weight of the LCU; lambda is calculated; and a QP (Quantization Parameter) is calculated according to the lambda. The bit rate allocation is carried out by taking the visual saliency as the weight, so a PSNR of a saliency region is improved, the subjective quality of a reconstructed video saliency area is relatively good, and a bit rate error is relatively low. The PSNR of the saliency region is improved. The subjective quality of the saliency region is improved. The PSNR of a non-saliency region is reduced a little. The attention degree of eyes for a non-region of interest is relatively low, so the change of the subjective quality of the non-saliency region can be ignored.

The invention relates to a monitoring video compression and reconstruction method based on a visual attention mechanism. The method comprises the following steps: dividing a monitoring video into multiple video frame groups, wherein each video frame group determines a key frame; determining an interested region in each video; compressing and reconstructing the key frame and the non-key frame, wherein the step of compressing and reconstructing the non-key frame comprises steps of obtaining image residual obtained by subtracting pixel values of the non-key frame and the corresponding reconstructed key frame, performing blocked compression of different degrees according to the interested region and the non-interested region, reconstructing the compressed blocks of the image residuals at different regions by using different algorithms, and then adding the reconstructed compressed block with the corresponding reconstructed key frame to obtain the reconstructed non-key frame; and finally arranging and reconstructing the reconstructed key frame and the reconstructed non-key frame according to the order before reconstruction, thereby obtaining the monitoring video. The monitoring video iscompressed in different measurement ratio based on the residual application visual attention mechanism blocking, the reconstruction effect is guaranteed, and the compression ratio is reduced.

The invention discloses a real-time detection method of track occupying phenomenon based on digital image processing, which comprises the following steps: selecting a background image and converting the background image into a gray scale image; extracting video frames from the real-time monitoring video and buffering the video frames; The corridor of the gray image of the background image is takenas the region of interest and the non-region of interest of the background image is shielded by a mask. The background image is used to extract the foreground of the video frame without updating theGaussian mixture model, and the binary foreground image is saved. The binary foreground image is detected and fused to obtain the fused foreground image. denoising is performed on the image after theforeground fusion to obtain a denoised binary foreground map; The denoised binary foreground map obtained in the previous step is labeled in the real-time image frame to predict and warn. the method can make real-time and accurate positioning of the existence of road occupying information, liberate the labor force, reduce the input of personnel, and increase work efficiency.