51results about How to "Preserve edge detail" patented technology

A method for restoring hyperspectral remote sensing image based on non-convex and low-rank sparse constraint belongs to the field of hyperspectral remote sensing image processing in remote sensing image processing. In order to solve the problem that the existing hyperspectral remote sensing image restoration technology can not effectively remove noise and improve the image restoration quality, themethod comprises the following steps: inputting a hyperspectral remote sensing image; initializing a weight coefficient matrix, iterative times and a convergence threshold, initializing sub-image size and scanning step, partitioning sub-blocks; establishing an image restoration model; the auxiliary variable and the coefficient of the regular term being introduced, and the maximum-minimum algorithm being used to solve the problem iteratively; judging whether the restoration result satisfies the convergence condition; obtaining a hyperspectral restored image that meets the requirements by iterative times, otherwise returning to corresponding steps to continue the iterative operation; calculating a weight coefficient matrix and assigning appropriate weights to each sub-block; hyperspectral remote sensing images being restored to obtain the final restored hyperspectral remote sensing images. The effect of denoising is obvious and the image details are preserved.

The invention relates to a mathematical morphology-based image edge detection method. The conventional method has defects in aspects of inhibition of noise and fake edges. The method has the following steps of: performing morphological processing of an image by using structuring elements of various scales to obtain an edge information image; performing weighted coalition of the edge information image to obtain a total edge information image; and finally performing thinning and thresholding segmentation of the total edge information image to obtain a final binary edge image. In the method, the structuring elements of various scales and the improved edge detection operators are adopted, which not only effectively filters noise and inhibits fake edges, but also keeps edge details to the maximum extent, so that the good edge effect is achieved.

The invention discloses a super-resolution reconstruction method of multi-frame images. A steering kernel regression method is introduced separately to a clustering phase of an adaptive sparse learning model and a regularized reconstruction process; the two methods are enabled to complement each other's advantages and disadvantages, forming a reconstruction algorithm of better performance. Motionregistration is not required, and therefore, reconstruction errors due to registration errors are avoided and the quality of a reconstructed image can be significantly improved; the method is applicable to places including any motion mode. Compared with existing learning-based multi-frame super-resolution reconstruction methods, the simple efficient integrated reconstruction model that integratessteering kernel regression and sparse learning is provided which can make use of both global structural self-similar priori constraints and sparse constraints to carry out regressive estimation, so that edge details of an image can be better maintained, and image high-frequency distortion can be relieved; the method is not limited to motion amplitude or application scenes and accordingly is adaptive to complex environments.

The invention discloses a thresholding image denoising method based on nonsubsampled Contourlet transformation, and the problem that the effect of an existing denoising method is bad is mainly solved. The realization method comprises the steps of inputting a noisy image, conducting the nonsubsampled Contourlet transformation on the noisy image, conducting a sub-band noise variance estimation and an estimation of a standard error of a coefficient in a transform domain, setting a reasonable threshold, conducting threshold processing on an NSCT coefficient through an improved semi-soft threshold function without processing a low frequent component, and obtaining a de-noised image through conducting nonsubsampled Contourlet inverse transformation on a processed NSCT coefficient. According to the thresholding image denoising method based on the nonsubsampled Contourlet transformation, noise in a natural image with white Gaussian noise can be effectively removed, and image information edges are kept as much as possible at the same time, so that the best recovery of an original image is obtained, and the thresholding image denoising method based on the nonsubsampled Contourlet transformation has good application prospect and great development potential.

The invention discloses a robust roundabout detection method for a lunar probe and an aircraft navigation method, solves the problems that a conventional roundabout detection method is influenced by illumination and is poor in robustness in the descending and landing process of the existing lunar probe, and belongs to the technical field of lunar detection. According to the method, firstly, an image is filtered, image edge details are reserved, and noise influence is removed; secondly, entropy distribution of the processed image is calculated, maximum entropy threshold segmentation is carried out, image information is subjected to ternary processing, the sensitivity of the image to a light source is removed, and meanwhile the image information is reserved to the maximum degree; according to the method, a normalized multi-index constraint ring-shaped mountain matching and fitting method is adopted to complete ring-shaped mountain extraction, and continuous, complete and clear edges of the lunar surface ring-shaped mountain can be extracted under different illumination conditions. The aircraft navigation method based on robust annular mountain extraction is better in real-time performance.

The invention discloses an image edge processing method based on guided filtering and application, and relates to the technical field of digital image processing. The method comprises the following steps of: acquiring an input image and a guide image; collecting a set first rectangular window wk1 used for weak denoising and a set second rectangular window wk2 used for strong denoising; using the window wk1 for guided filtering weak denoising, and using the window wk2 for guided filtering strong denoising; obtaining a denoising intensity absolute difference according to the filtering coefficients corresponding to the two windows, and obtaining a corresponding filtering weight according to the absolute difference; obtaining the value of the edge gradient of a strong denoising result image, and then judging pixel information needing edge smoothing processing; and based on a weak denoising result image and the filtering weight, carrying out edge smoothing processing on pixels to obtain a final result image. According to the method, edge smoothing can be achieved while image noise is suppressed and edge details are kept, the denoising intensity and the edge smoothing intensity can be independently controlled, the applicability is wide, and the flexibility is high.

The invention discloses an image stereo matching method and a binocular vision device. The method comprises steps: stereo matching cost calculation is carried out on a reference pixel point in a firstphotographed image and multiple to-be-matched pixel points in a second photographed image to determine a matching cost value of the reference pixel point and each to-be-matched pixel point, wherein the first image and the second image are images photographed by different cameras of the binocular vision device towards the same scene; a matching point for the reference pixel point is screened fromthe multiple to-be-matched pixel points; and whether the screened matching point is stable is judged according to a matching cost curve, an instable matching point is replaced, and a disparity map isobtained. The technical scheme can effectively improve foreground expansion effects of the disparity map, the operation efficiency is improved, instable matching points are removed through adopting amedian filter method, the disparity map precision can be enhanced, and edge details of the disparity map can also be kept.

The invention discloses a pest image recognition method based on a Bayesian width residual neural network. The method comprises the following steps: step 1, preprocessing the pest image recognition training data set, using a rich edge detection algorithm (Rich-Edge) Pest edge detection is performed on the grayscale image. Step 2. Construct a Bayesian Wide-Residual Network (BWResNet). Step 3: Input the pest edge image obtained in Step 1 into the BWResNet constructed in Step 2. Use the pest edge image training set obtained in step 1 to obtain the total error function of BWResNet. Step 4: Use the error function obtained in Step 3 to train the network. Here we propose the block-conjugate (Block‑cg) algorithm to train the network. Step 5. Update hyperparameters according to the optimized network in step 4. Step 6: Repeat steps 4 and 5 to obtain the final network, and the classification accuracy rate obtained after inputting the verification set of pest images into the network is higher.

The invention discloses a thresholding image denoising method based on non-subsampling Contourlet transform, which mainly solves the problem of poor effect of the existing denoising methods. The implementation process is: 1) input a noisy image; 2) perform non-subsampling Contourlet transform on the noisy image; 3) estimate the noise variance of each subband and the coefficient standard deviation in the transform domain; 4) set reasonable 5) Threshold the NSCT coefficients using the improved semi-soft threshold function, and do not process the low-frequency components; 6) Perform non-subsampling Contourlet inverse transformation on the processed NSCT coefficients to obtain the denoised image. The invention can effectively remove the noise in the natural image containing Gaussian white noise, and while removing the noise, the edge of the image information is preserved as much as possible, so as to obtain the best restoration of the original image, and has a good application prospect and great development. potential.