269 results about "Gaussian pyramid" patented technology

A method detects an object, such a face, in an image. The image is first partitioned into patches of various sizes using either an integral image or a Gaussian pyramid. Features in each patch are evaluated to determine a cumulative score. The evaluating is repeated while the cumulative score is within a range of an acceptance threshold and a rejection threshold, and otherwise the image is rejected when the accumulated score is less than the rejection threshold and accepted as including the object when the cumulative score is greater than the acceptance threshold.

A method of automatically detecting and tracking successive frames in a region of interesting by an electronic imaging device includes: decomposing a frame into intensity, color and direction features according to human perceptions; filtering an input image by a Gaussian pyramid to obtain levels of pyramid representations by down sampling; calculating the features of pyramid representations; using a linear center-surround operator similar to a biological perception to expedite the calculation of a mean value of the peripheral region; using the difference of each feature between a small central region and the peripheral region as a measured value; overlaying the pyramid feature maps to obtain a conspicuity map and unify the conspicuity maps of the three features; obtaining a saliency map of the frames by linear combination; and using the saliency map for a segmentation to mark an interesting region of a frame in the large region of the conspicuity maps.

A method and system for dual energy image registration is disclosed. In order to segment first and second images of a dual energy image pair, the first and second images are preprocessed to detect edges in the images. Gaussian pyramids, having multiple pyramid images corresponding to multiple pyramid levels, are generated for the first and second images. An initial optical flow value is initialized for a first pyramid level, and the optical flow value is sequentially updated for each pyramid level based on the corresponding pyramid images using an optimization function having a similarity measure and a regularizer. This results in a final optical flow value between the first and second images, and the first and second images are registered based on the final optical flow value.

The invention provides a remote sensing image scene classification method based on multi-scale depth feature fusion and transfer learning in order to solve the classification problem of remote sensingimage scenes. The method comprises the following steps: firstly, obtaining a multi-scale remote sensing image by using a Gaussian pyramid algorithm, inputting the multi-scale remote sensing image into a full convolutional neural network, and extracting multi-scale depth local features; cutting the image to a fixed size required by the CNN; obtaining global features of a full connection layer in an input network, using compact bilinear pooling operation to encode multi-scale depth local features and global features obtained by a CNN, fusing the two depth features to jointly represent a remotesensing image, enhancing the mutual relation between the features, and enabling the obtained features to be more distinctive; and finally, classifying the remote sensing image scenes by utilizing a transfer learning technology and combining the two methods. According to the convolutional neural network, a VGG16-Net is used as a basic network.

The disclosure relates to an image completion method and device. The method includes the following steps: generating n layers of Gauss pyramids on the basis of an image to be restored, n being a positive integer; through a block matching method, completing images of each of the n layers of Gauss pyramids, and obtaining an actual image of each layer after completion; generating a generative network of an adversarial network on the basis of the n image of the n layers of Gauss pyramids and the actual images of all the layers; and on the basis of the n image and the generative network, completing the image to be restored, and obtaining a completed image after completion. According to the technical scheme, the generative network is obtained through a Gauss pyramid of the image to be restored, the actual images of all layers of Gauss pyramids and the adversarial network, so a clear and complete image high in resolution can be generated quickly. Therefore, the resolution of the completed image is improved, learning about the generative network is accelerated, and a more real image can be obtained.

The invention belongs to the technical field of image salient region detection and particularly discloses an image salient region detection method based on joint sparse multi-scale fusion. The image salient region detection method comprises the following steps: (1) constructing a multilayer Gaussian pyramid for a training image to realize multi scales and training to obtain a dictionary under each scale; (2) obtaining an image block of each pixel point in a test image and carrying out joint sparse solution of a sparse representation coefficient of the image block under each scale; (3) taking the sparse representation coefficient as a feature to carry out saliency calculation; (4) fusing salient results under the multi sales to obtain a final salient image. The image salient region detection method has the benefits that the purpose of extracting a region capable of catching people's eyes in any given image is realized; the image salient region detection method has the advantages that firstly, the effect under different image scales is overcome under multi-scale operation; secondly, a joint sparse framework is very beneficial to saliency calculation; experiments show that the results obtained by the method have better robustness and are inferior to those obtained according to most of the conventional methods.

The invention discloses an automatic fast segmenting method of a tumor pathological image. The method comprises the following steps: firstly filtering a tumor original pathological image through the adoption of a Gaussian pyramid algorithm to respectively obtain pathological images with equal resolution, double resolution, fourfold resolution, eightfold resolution and 16-fold resolution; determining an initial region of interest containing the tumor on the equal resolution image through a RGB color model and morphological close operation; iteratively optimizing the initial regions of interest from the equal resolution to the fourfold resolution through the adoption of bhattacharyya distance; judging that the contribution of the RGB color model to the tumor region of interest has been reduced to zero when the bhattacharyya distance achieves a set threshold value; performing the self-adaptive high resolution selection of the deep precise segmentation through the adoption of a convergence exponent filtering algorithm, thereby further segmenting under the most suitable high resolution; and finally segmenting out a normal tissue and a tumor tissue in the tumor region of interest through the adoption of a bag of words model based on random projection. The method disclosed by the invention has the features of being accurate, fast and automatic.

The invention discloses a single-exposure high-dynamic range image generation method adapted to a medical image. The method comprises the steps that 1) images of different exposure are generated on the basis of reverse tone mapping, and a reverse tone mapping function is modeled; 2) according to a pseudo-exposure HDR image generated in the first step, a weight graph of the image is obtained on the basis of a pyramid decomposed multi-exposure image fusion algorithm; 3) Laplacian pyramid decomposition is carried out on a sequence of N multi-exposure images, Gaussian pyramid decomposition is carried out on N weight graphs, and images and weight graphs of different resolutions are obtained, and a fusion formula is obtained; and finally, Laplacian inverse transformation is carried out on a pyramid L{R}l to obtain a fused image R. Thus, an HDR image synthesized by input image is obtained. Compared with an original image, the synthesized HDR image is higher in contrast, more distinct in colors, and more reinforced in details; a fusion algorithm is low in computational complexity; and high dynamic range image generation can be realized in real time.

The invention discloses a remote sensing image region of interest detection method based on integer wavelets and visual features, which belongs to the technical field of remote sensing image target identification. The implementing process of the method comprises the following steps: 1, performing color synthesis and filtering and noise reduction preprocessing on a remote sensing image; 2, converting the preprocessed RGB spatial remote sensing image into a CIE Lab color space to obtain a brightness and color feature map, and converting an L component by using integer wavelets to obtain a direction feature map; 3, constructing a Gaussian difference filter for simulating the retina receptive field of a human eye, performing cross-scale combination in combination with a Gaussian pyramid to obtain a brightness and color feature saliency map, and performing wavelet coefficient sieving and cross-scale combination to obtain a direction feature saliency map; 4, synthesizing a main saliency map by using a feature competitive strategy; and 5, partitioning the threshold values of the main saliency map to obtain a region of interest. Due to the adoption of the remote sensing image region of interest detection method, the detection accuracy of a remote sensing image region of interest is increased, and the computation complexity is lowered; and the remote sensing image region of interest detection method can be applied to the fields of environmental monitoring, urban planning, forestry investigation and the like.

A method, system and computer program product for representing an image is provided. The image that needs to be represented is represented in the form of a Gaussian pyramid which is a scale-space representation of the image and includes several pyramid images. The feature points in the pyramid images are identified and a specified number of feature points are selected. The orientations of the selected feature points are obtained by using a set of orientation calculating algorithms. A patch is extracted around the feature point in the pyramid images based on the orientations of the feature point and the sampling factor of the pyramid image. The boundary patches in lo the pyramid images are extracted by padding the pyramid images with extra pixels. The feature vectors of the extracted patches are defined. These feature vectors are normalized so that the components in the feature vectors are less than a threshold.

The invention discloses a detection method and a detection system for video image text and relates to the technical field of image text detection. The method comprises the following steps of coarse positioning of video image text areas and verification of video image text areas. The specific steps are as follows: (A) coarse positioning of video image text area: the gaussian pyramid multi-scale image of a video image is calculated, the image text area is segmented by threshold segmentation, text areas are combined by the inflation and corrosion technology in mathematical morphology, and candidate text areas are analyzed and positioned by connected domain analysis, text line segmentation and elimination rules; and (B) verification of video image text areas: an adaboost classifier is trained by adopting harr characteristics, and the adaboost classifier is utilized to detect whether text exists in the candidate text areas and the candidate text areas the text of which is not detected are eliminated. The method and the system can improve the accuracy and the recall ratio of video image text detection.

The invention discloses a target tracking template updating method. The method comprises the following steps: determining whether a target tracking template needs to be updated; extracting feature points of a target region in the last frame and a current frame of an image, matching the feature points of the target region of the last frame with the feature points of the target region of the current frame, and re-determining the position of a tracked target; using the feature points of the target region of the current frame as new feature points to carry out tracking; and using the target region of the current frame as an updated target tracking template. The beneficial effects are as follows: setting of determination conditions of target tracking template updating avoids the situation that the template cannot catch up with the changed target and prevents noise accumulation during tracking, so that the target tracking is continuous and stable; establishment of an image Gauss pyramid improves the update efficiency; a quick feature point detection method enhances the repeatability of feature detection and prevents unstable edge responses; and the method is suitable for tracking a changing target, especially when wide-angle overturn happens.

Provided are an STFT parallelization system and method based on recursion Gaussian filtering on a CUDA platform. The method comprises the steps that first, original images are transmitted to a GPU end for conducting a series of Gaussian filtering and downsampling to establish a Gaussian pyramid, Gaussian filtering is conducted through a recursion Gaussian filter, and then substraction is conducted on the adjacent images to obtain a Gaussian difference pyramid; second, a thread block is used as a unit to load in an image, each thread is used for processing one pixel, and the pixel is compared with the adjacent 26 pixels to obtain local extreme points; third, each thread is used for processing one local extreme point, and positioning and selecting of key points are conducted; fourth, one thread block is used for calculating the direction of one key point, one thread is used for calculating the direction and the amplitude value of one pixel in the neighbourhood of the key point, the direction and the amplitude valve are accumulated to a gradient histogram through an atomic addition provided by a CUDA, and the information such as the coordinates and the directions of the key points are transmitted to a host end according to the directions of the key points obtained by the gradient histogram; fifth, one thread block is used for calculating one key point descriptor, then a calculating result is transmitted to the host end, and SIFT feature extraction is achieved. The STFT parallelization system and method based on the recursion Gaussian filtering on the CUDA platform improve the calculating speed of an SIFT algorithm.

The invention discloses a multi-scale detail fusion multi-exposure high dynamic image reconstruction method, and relates to the field of image processing technology. The method includes the steps of taking into account three measurement factors of image contrast, saturation, and moderate exposure, calculating a weight map of an original multi-exposure image according to the measurement factors, and generating a weight Gaussian pyramid and a Laplacian coefficient pyramid by the use of the pyramid principle; constructing a weight map mapping function, and performing weight mapping on the Gaussian pyramid; then according to the pyramid principle, performing weighted averaging on the weight map and a Laplacian coefficient of each layer to obtain an integrated Laplacian pyramid; and finally, obtaining an HDR image by inverse transformation of the integrated pyramid. The qualitative and quantitative analysis of the fused image effect is performed. Compared with other multi-exposure image fusion methods, the method of the invention effectively preserves the details of a low dark region and a highlight region, has a significant fusion effect and can better reflect the details of a real scene.

The invention discloses a frequency-domain-analysis-based method for detecting a region-of-interest of a visible light remote sensing image, belonging to the technical field of remote sensing image processing and image recognition. The method comprises the following implementation processes of: (1) preprocessing the image; (2) carrying out quaternion Fourier transformation on the preprocessed result to obtain the frequency domain information of the image; (3) retaining a phase spectrum and obtaining the high-frequency information of a magnitude spectrum by using a Butterworth high-pass filter; (4) carrying out quaternion Fourier inversion on the phase spectrum and the high-frequency information of the magnitude spectrum to obtain a characteristic pattern; and (5) filtering the characteristic pattern by using a gaussian pyramid and reducing dimensions to obtain a saliency map; (6) carrying out threshold segmentation and binaryzation on the saliency map to obtain a region-of-interest template; and (7) raising the dimensions of the template and carrying out masking operation on an original image so as to obtain the final region-of-interest. The method realizes the rapid and accurate location of the region-of-interest, has the advantages of high accuracy of region description, low computation complexity and the like and can be applied in fields such as environmental monitoring, land utilization, agricultural investigation and the like.

The invention provides a local image describing method based on an RGB-D sensor. The local image describing method comprises the following steps: calibrating parameters of the RGB-D sensor, and carrying out image pretreatment and local characteristic extraction on acquired RGB images and depth images; expressing the central point of the extracted local characteristics through three variables of space partitioning, grayness sequence marking and normal vector sequence marking; calculating out the scale value and the principal direction of characteristic points according to the data of the depth images in the central point of the local characteristics; building a three-dimensional histogram, unfolding the three-dimensional histogram and normalizing the three-dimensional histogram to be a one-dimensional vector which is a descriptor for the local characteristics. Compared with two-dimensional, three-dimensional and fusion descriptors, the method has the advantages of obvious lighting robustness and the scene changeable robustness. Based on depth scale invariance and rotating invariance, in the method provided by the invention, point cloud depth data is used for replacing estimation to the scale space of a Gauss pyramid, so that the speed is accelerated greatly.

The invention relates to a method for image mosaic based on a feature detection operator of second order difference of Gaussian, which carries out mosaic on sequence images which vary in viewpoint, rotation, proportion, illumination and the like to a certain degree. The method provided by the invention adopts zero crossing point detection of second order difference of Gaussian (D<2>oG) pyramid to replace extreme point detection of the original difference of Gaussian (DoG) pyramid so as to extract scale invariant feature points, thereby effectively simplifying the structure of the Gaussian pyramid. The method comprises the steps of: first, extracting image feature points by using an improved SIFT (Scale Invariant Feature Transform) algorithm; then, searching a rough matching point pairs for the extracted feature points through a BBF (Best-Bin-First) algorithm, and purifying the feature point matching pairs by adopting an RANSAC algorithm so as to calculate an invariant transformation matrix H; and finally, completing seamless mosaic for the images by adopting a fading-in-and-out smoothing algorithm. Experimental results show that the method improves the accuracy and the real-time performance of image mosaic, can well solve problems such as illumination, rotation, scale variation, affine and the like, and realizes automatic mosaic without manual intervention.

The invention discloses an SAR image target fast detecting method based on significance detecting and clustering and mainly solves the problem that existing SAR image target detecting is slow in detecting and high in clutter false alarm rate. The method includes: building an intensity channel Gaussian pyramid for an SAR image A; selecting the dimension of the Gaussian pyramid according to a target size; building an intensity significance image for the SAR image under the selected dimension according to the selected dimension; sequentially determining the most significant areas on the intensity significance image, and acquiring a binary image marking all significant areas; performing morphological filter and clustering on the binary image, and then performing area removing according to area range of the target area; using the area-removed clustered binary image to perform suspected target slice extraction to complete SAR image target detection. The method is fast in detection, low in false alarm rate and suitable for SAR image target fast detection.

The invention discloses a weighted fusion-based underwater image enhancing method. The underwater image enhancing method comprises the following steps of subjecting a degraded underwater image to the equalization treatment by using Gray-World and histogram respectively so as to obtain an input image; through the normalization treatment, defining a weight factor through the weighted guide filtering method, and correcting the weight factor to obtain a corrected weight graph; decomposing each input image respectively through a laplacian pyramid, and decomposing each weight graph through a gaussian pyramid, and finally fusing the input image and the weight graph through the multi-scale fusion method to obtain an image rich in details. Based on the method, the complicated deconvolution operation is not required, and weights are reasonably selected. Therefore, the image is richer in details on the basis of color correction.

A method, system and computer program product for representing an image is provided. The image that needs to be represented is represented in the form of a Gaussian pyramid which is a scale-space representation of the image and includes several pyramid images. The feature points in the pyramid images are identified and a specified number of feature points are selected. The orientations of the selected feature points are obtained by using a set of orientation calculating algorithms. A patch is extracted around the feature point in the pyramid images based on the orientations of the feature point and the sampling factor of the pyramid image. The boundary patches in the pyramid images are extracted by padding the pyramid images with extra pixels. The feature vectors of the extracted patches are defined. These feature vectors are normalized so that the components in the feature vectors are less than a threshold.

The invention discloses an intelligent transportation moving object detection tracking method which is used for detecting and tracking a moving object in an intelligent transportation video. The intelligent transportation moving object detection tracking method comprises the following steps: considering objects with obvious edge characteristics such as guard bars in the middle part and on the two sides of a road in a video frame image; adopting a finite difference first-order partial derivative matrix to calculate the gradients of an input front video frame image and a back video frame image, and extracting a fine outline of an input image characteristic area; performing outline connection, area filling consolidation and morphological processing to obtain a preprocessed image; adopting an improved SIFT algorithm to extract key points from the preprocessed image, and only establishing a six-layered Gaussian pyramid for the preprocessed image, so as to ensure the situation that a few of key points can be detected when the image is changed in a small scale, meanwhile reduce redundant calculation requirement on establishing the pyramid, and improve the time efficiency of the algorithm; before frame difference, the image of the video sequence front frame and back frame is rectified by utilizing an affine transformation model obtained by the preprocessed image, so as to eliminate the influence of the video frame shift on performance detection.

The invention discloses a video target tracking method, in particular a particle filter and light stream vector-based video target tracking method. The method comprises the following steps of: first, establishing an L-level Gaussian pyramid for an image at the moment t, and calculating the gray gradient of each level of Gaussian pyramid image along directions x and y; then, solving position deflection of M characteristic points of the image at the moment t-1 in the image at the moment t; next, resampling N particles for the image at the moment t-1 and redistributing the newly sampled N particles by using a random number or a light stream vector; later on, recalculating an RGB histogram of the N particles, calculating the weight of each particle according to the RGB histogram and performing weighted averaging for the N particle positions according to the weight to obtain an estimated position for the image of a tracking object at the moment t; and finally, performing stability comparison calculation on the estimated position to obtain a final position of the tracking object. The method can efficiently and accurately track a video target.

The invention discloses an extraction method of visual salient regions based on multiscale relative entropy. The extraction method of the visual salient regions based on the multiscale relative entropy comprises extracting different global color feature characteristic patterns and direction characteristic patterns from input images, carrying out gaussian pyramid decomposition and multiscale normalization on the global color feature characteristic patterns and the direction characteristic patterns, respectively computing color partial relative entropy and direction partial relative entropy of respective feature space, respectively normalizing results of the different feature space, carrying out linearity superposition of the results of the different feature space, linearly adding color saliency maps and direction saliency maps together, carrying out two-dimensional gaussian smoothing, and therefore extracting visual salient regions. Compared with a traditional method, the extraction method of the visual salient regions based on the multiscale relative entropy takes a full account of color features and direction features reflecting global significance and relative entropy reflecting partial novelty, adopts a method of multiscale analysis, and has the advantages of being effective and reliable. The extraction method of the visual salient regions based on the multiscale relative entropy is capable of being used for self-adaption compressed sampling and target detection of natural images, remote sensing images and the like, and is wide in application prospect of lowcost imaging devices.

The invention relates to an infrared image vehicle detection method based on auxiliary road information and significance detection. According to the method, at first, a gauss pyramid is used for carrying out background evaluation on a source image, and high-frequency information obtained by subtracting a background evaluation image from the source image comprises targets and background clutter; OTSU threshold segmentation is carried out on the source image to obtain a binary image to serve as estimation of the road information; an interest area can be obtained through collation operation of the high-frequency information and the road information; the significance detection is carried out on the center and the peripheral area of the interest area; track relevance is carried out on the candidate targets, and therefore the vehicle targets can be well extracted.

The invention discloses a mammogram mass detection system and detection method based on a visual attention mechanism, which mainly solves the problems of low detection rate and high false positive rate in the prior art. The whole system includes: image preprocessing module, feature extraction module, tumor detection module and detection result output module. The image preprocessing module preprocesses the original image; the feature extraction module extracts eigenvalues ​​from the preprocessed image; the mass detection module first generates a Gaussian pyramid of the feature map for the eigenvalues; and then processes the Gaussian pyramid to obtain a total saliency map; Finally, the total saliency map is segmented to obtain candidate suspicious masses and filter out false positive masses; the detection results are output through the detection result output module. The invention has the advantages of high detection rate and low false positive rate for breast X-ray image masses, and can be used for detection of interest regions of medical images and its auxiliary diagnosis.

The invention discloses a nonuniform sparse sampling video super resolution method, which belongs to the technical field of video super resolution. The method comprises key technologies such as shot detection, key frame extraction, original image fuzzy processing and down sampling, nonuniform sampling vision model building, sparse dictionary construction, and video reconstruction. A nonuniform sparse sampling method based on foveated vision is adopted, differential sparse sampling is carried out on a video sequence, the number of atoms is greatly reduced, a dual dictionary of high and low resolutions (high and low resolution reference copy) is generated in real time or calculated in a prior mode according to the resolution of a mobile terminal device screen as carried meta data, a hardware-supported nonlinear Mipmap interpolation method is used for simulating and generating a Foveation image in real time, a result similar to that in a Gauss pyramid method can be acquired, and the computation overhead is lower.

The invention discloses a remote-sensing image building detection method based on multi-scale and multi-characteristic fusion. The method comprises steps that through high resolution remote-sensing image downsampling, an image pyramid formed by images in different scales is acquired; edge images of the image pyramid are calculated; a characteristic model is established through multi-set characteristic calculation and fusion of the edge images in different scales; according to the characteristic model and neighborhood local non-maximum inhibition, window selection is carried out to acquire a target window; small-scope expansion / contraction calculation of the target window is carried out to acquire a rectangular window; the rectangular window is turned according to a main direction of the target window to acquire an optimal target window, and a building is extracted according to the optimal target window. The method is advantaged in that multi-scale building detection on the Gaussian pyramid image is carried out, universality of detection on buildings different in sizes, shapes and directions is realized, and automatic building detection precision and efficiency are effectively improved.

The embodiment of the invention discloses an image processing method and device based on a Laplace pyramid, wherein the image processing method comprises the steps of: constructing a Gaussian pyramid corresponding to an input image; generating an intermediate image according to the Gaussian pyramid and a mapping function r(x); obtaining a coefficient of the Laplace pyramid according to the intermediate image; correspondingly outputting the coefficient of the Laplace pyramid based on the intermediate image to the Laplace pyramid until each point of the Laplace pyramid is filled, obtaining a new Laplace pyramid; and reconstructing an image according to the new Laplace pyramid, and obtaining an output image. By implementing the embodiment of the invention, the restraint that the traditional Laplace pyramid can not be used for carrying out edge preserving image enhancement, the edge preserving image enhancement is carried out by using the Laplace pyramid, the image processing method more simply and flexibly realizes image enhancement in comparison with the traditional edge preserving image enhancement methods (such as two-side filtering and wavelet methods), the time complexity can be reduced, and additional excessive parameters are not needed to be set.

The invention relates to a multi-exposure image fusion method, which solves the technical problems of incomplete detail information retention and incomplete ghosting elimination, and comprises the following steps of: 1, constructing an initial weight map by using exposure brightness and chroma information of a multi-exposure image sequence; 2, firstly, moving object detection is conducted on the low-dynamic image sequence, a moving area is calculated, and then a ghosting eliminating method based on super-pixel segmentation is used for eliminating ghosting; 3, completing ghosting correction onthe initial weight map in the step 1; 4, performing normalization processing on the weight map after ghosting correction in the step 3; and 5, constructing a weighted Gaussian pyramid according to theweight graph in the step 4, constructing a Laplacian pyramid of a low-dynamic image sequence, defining detail gain items, calculating a fusion pyramid, and performing image reconstruction according to the fusion pyramid to obtain a fusion image HDR. The problem is well solved, and the method can be used for image processing.

In order to solve the problems of long operating time and low matching ratio of an SIFT (Scale Invariant Feature Transform) algorithm, the present invention puts forward an improved SIFT algorithm. On the basis of the original classic SIFT algorithm, a two-dimensional Mallat fast wavelet transformation algorithm is introduced, a low frequency component of an image is reconstructed, then the group number of Gauss pyramids is adjusted, downsampling frequency is reduced, and finally mismatching points are rejected through an optimized RANSAC algorithm. The improved SIFT algorithm of the present invention not only reduces the consumed time of matching, but also improves the matching rate, thus the improved SIFT algorithm is better than the original SIFT algorithm.