154 results about "Silhouette edge" patented technology

A method detects silhouette edges in images. An ambient image is acquired of a scene with ambient light. A set of illuminated images is also acquired of the scene. Each illuminated image is acquired with a different light source illuminating the scene. The ambient image is combined with the set of illuminated to detect cast shadows, and silhouette edge pixels are located from the cast shadows.

A covering panel has a support plate, made from synthetic plastic material, with a flat and even top surface, a bottom face and pairs of opposite sides provided with complementary profile edges in the form of a male profile edge on one side and a female profile edge on the other, where said male profiled edge can be connected to a female edge of an equal adjacent covering panel, preferably according to a direction that runs perpendicular to said top surface. In their bottom face there are strip-shaped recesses that run according to the direction of extrusion and alternate with a number of support strips, of which at least a part has a width S along that bottom face that complies with 1.5 T<S<10 T, and whereby 0.4 T<C<0.7 T applies, in which C is the average depth of these recesses. A process extrudes a synthetic plastic plate.

The invention discloses a method for rendering contour edges of models. The method includes drawing a model to be contoured to a preset rendering target; converting coordinates of a model bounding box, copying the preset rendering target and corresponding projection coverage range to a contouring bitmap; determining texture coordinates of the projection coverage range to generate a planar rectangle vertex data stream according to attributes of a background buffer area and the projection coverage range; determining contour edges of the model on the surface of an original rendering target according to a preset template of the background buffer area, and rendering the contour edges according to the preset scheme. By the method for rendering the contour edges of the models, the background buffer area is processed locally by the aid of the bounding box, the color, the depth and the template in the pixel space, load to a central processing unit and a diagram processor is reduced, varieties of rendering of the contour edges are developed to the greatest extent, extendability of rendering effect is improved and rendering efficiency is improved.

A method generates an image with de-emphasized textures. Each pixel in the image is classified as either a silhouette edge pixel, a texture edge pixels, or a featureless pixel. A mask image M(x, y) is generated, wherein an intensity of a given pixel (x, y) in the mask image M(x, y) is zero if the pixel (x, y) is classified as the texture edge pixel, is d(x, y) if the pixel (x, y) is classified as the featureless pixel, and is one if the pixel (x, y) is classified as the silhouette edge pixel. An intensity gradient ∇I(x, y) is determined in the masked image, and the intensity gradients in the masked image are integrated according to G(x, y)=∇I(x, y). M(x, y). Then, an output image I′ is generated by minimizing |∇I′−G|, and normalizing the intensities in the output image I′.

The invention discloses a target recognition method based on an image contour characteristic. The method comprises the steps of preprocessing a template image and a to-be-identified object image for generating a binary image; establishing a characteristic database of an object template contour; extracting a whole contour of an object template binary image, equidistantly acquiring a certain number of characteristic points on the contour, and describing the object contour by means of a context characteristic of the characteristic points; performing target recognition on the to-be-identified object image; extracting the contour edge of the to-be-identified binary image; selecting a certain number of characteristic points; converting the shape direction of the to-be-identified object to the direction of the template; describing the contour of the to-be-identified image after direction changing by means of the context characteristic of the selected points; and determining similarity between the to-be-identified object and the template object through matching cost. Compared with prior art, the target recognition method has advantages of settling a problem of rotational invariance in a contour matching process, realizing invariable rotation in the contour matching process, and realizing effective application in target recognition in the image.

The present invention discloses a flexible robot vision recognition and positioning system based on the depth learning. The system is implemented in the following steps: obtaining an image of a part, carrying out binarization processing on the image of the part to extract an outer contour of the image of the part; finding out a circumscribed rectangle of the outer contour edge in the lateral direction, determining to-be-recognized areas, and normalizing the areas to a standard image; gradually rotating the standard image at an equal angle, finding out a rotation angle alpha when the standard image is rotated to a minimum area of the circumscribed rectangle of the outer contour edge in the lateral direction; using a depth learning network to extract the outer contour edge when the rotation angle is alpha, and recognizing the part and the pose of the part; and according to the rotation angle alpha and the pose, calculating the actual pose of the to-be-recognized part before rotating, and transmitting the pose data to a flexible robot, so that the flexible robot can pick up the to-be-recognized part. According to the system disclosed by the present invention, contour shape features contained in the part image data are automatically extracted layer by layer by using the depth learning network, so that accuracy and adaptability of part recognition and positioning can be greatly improved under the complicated conditions.

The invention discloses a human detection method. The human detection method includes the following steps that gradient histogram features are extracted in the depth color direction by the cooperation of color information with depth information; detection models are trained; region segmentation is conducted on test images, and the images are detected based on a segmentation result to obtain a candidate region (SR); the candidate region (SR) is verified based on edge detection. The human detection method based on color information and depth information is achieved, the obtained features comprise color information and depth information by the cooperation of feature extraction, outline edge information is enhanced, and then the features are more representative. Detection strategies based on different features and different classification models are used for different parts, and the defect that accuracy of long-distance objects shot by a depth camera is low is overcome. The verification method based on edge information is adopted, and then the detection result is more accurate.

An image processing apparatus includes a contour-candidate-edge detecting unit that detects, as contour candidate edges, edges based on a gradient magnitude of each pixel in a target image; a contour-edge detecting unit that detect contour edges by performing thresholding on gradient magnitudes of the contour candidate edges; an interpolation-line generating unit that generates interpolation lines for connecting end points of respective end-point pairs based on gradients of pixel values between the end points while each of the end point pairs is made up of an end point of an identical contour edge as a connection base and an end point of a different contour edge as a connection destination; and a contour-edge interpolating unit that selects one of the interpolation lines based on the gradients of pixel values on the interpolation lines and interpolates a contour edge between the end points with the selected interpolation line.

The invention discloses a self-adaption improved gradient information-based fruit surface defect detection method. The method comprises removing background of a RGB color image, carrying out binaryzation, individually extracting edge, carrying out expansion once to obtain a contour edge expansion image, converting the color image into a gray level image, carrying out calculation to obtain a normalized gradient image, carrying out statistics by a gradient histogram so that gradient information self-adaption improvement is realized, automatically calculating an image segmentation threshold value, acquiring an improved gradient binary image by image threshold segmentation, removing the contour edge expansion image of the improved gradient binary image to obtain a difference image, carrying out expansion hole-filling corrosion and median filtering treatment on the difference image to obtain a fruit surface defect image. The method realizes detection of surface different brightness characteristic defects under the condition of nonuniform globoid surface brightness, utilizes image segmentation threshold self-adaption calculation, is free of artificial selection, can be realized easily, and has an application potential in fruit and agricultural product quality machine vision on-line detection.

The invention belongs to the field of computer vision, and provides a method, a system and a computing device for generating a high-resolution depth map. The method and the system adopt an up-sampling algorithm to convert a low-resolution depth map into a high-resolution rough depth map, and then a skeleton map of the high-resolution rough depth map is further obtained. Then a low-resolution brightness map is utilized, and blocks which are most similar to each overlapping skeleton block in the skeleton map are searched in the low-resolution depth map. After that, a weighted stitching mode is adopted to fill the searched blocks into the corresponding positions of the high-resolution rough depth map so that the high-resolution depth map is obtained. As pixel information in the low-resolution depth map is supplemented to the high-resolution rough depth map, error depth data generated on the skeleton edge caused by data smoothing effect is corrected, a defect of simply applying the up-sampling mode is compensated and thus the obtained final high-resolution depth map is great in display effect. Besides, computing speed is fast.

The invention discloses a fast tracking recognition method for overlapped fruits by a picking robot. The fast tracking recognition method comprises the following steps: continuously collecting the latest ten frames of overlapped apple images through a camera; segmenting the collected first frame of image, and removing a background; determining the position of the circle center of overlapped apples by calculating the maximal value of the minimal distance from points in a circle to the edge of an outline; calculating the distance from the circle center to the edge of the outline to determine a radius; intercepting a subsequently matched template according to the circle center and the radius; determining the circle centers of overlapped apples in the continuously collected latest tens frames of images, and carrying out fitting and pre-judging on the motion path of the robot according to the circle center of each frame of image; determining the positions of overlapped apples in a next frame of image by synthesizing the radius and the pre-judging path, and intercepting the area of the overlapped apples; finally carrying out matching recognition by adopting a rapid normalized cross-correlation matching algorithm. According to the method, tracking recognition of near-spherical overlapped fruits such as the overlapped apples can be achieved; the running time is short; the picking efficiency of the picking robot can be effectively improved.

The invention discloses a polarized SAR (Synthetic Aperture Radar) image classification method based on sparse coding and a wavelet auto-encoder, mainly solving the problems of boundary classification caused by unreasonable characteristic extraction and poor region homogeneity caused by not considering spatial correlation. The method mainly comprises the steps of: (1) inputting images; (2) pre-processing; (3) extracting image characteristics; (4) sparse coding; (5) selecting a training sample and a test sample; (6) training a wavelet auto-encoder; (7) training a softmax classifier; (8) adjusting network parameters; (9) classifying the images; (10) coloring; and (11) outputting a classification result image. The method has a better denoising effect, considers data neighborhood information, and can better learn characteristics of a higher level from low-dimensional characteristics, allow a classification result image to have a clearer outline and edge, and improve polarized SAR image classification performance.

The invention relates to a method for picking up and identifying human ear characteristic by means of improved Hausdorff distance. The invention obtains standard human ear image through pretreatment of human ear image such as collection of human ear images, denoising of non-complexion noise, normalization of space dimension and illumination compensation; then, gray scale morphologic gradient and local threshold subdivision are adopted to pick up human ear edge feature so as to obtain standard human ear edge image. With the method and the Hausdorff distance improved through the length difference between standard variance and edge line segment, the influence of point set non-outline edge line segment point (outfield point) is reduced so as to obtain better anti-noise performance; moreover, the invention increases the accuracy of human ear edge image recognition by means of the characteristic value obtained on the basis of Hausdorff distance, thereby greatly increasing human ear recognition rate.

The invention provides a sub-pixel edge detection method based on improved morphology. The method comprises the steps that a digitized image of a product is obtained; morphology operators are applied for detecting the outline of the digitized image to obtain a pixel outline rough extraction region; Canny operators are adopted for detecting the whole pixel-level edge of the product from the pixel outline rough extraction region; by means of Gaussian edge functions obtained through ideal edge points and diffusion function convolution, the whole pixel-level edge is fitted into a sub-pixel-level edge of the product. According to the method, the edge detection operators of the morphology are improved, the edge of the image outline can be smoothed, edge details are kept better, anti-noise performance is improved, image edge information is kept, the smoothness and the continuity of the edge are kept, the image edge can be detected accurately, the connectivity of an original image is ensured, an image edge extraction region is reduced, and the processing speed is increased.

The invention relates to a machine vision detection method for a workpiece contour flange protrusion based on G-code guidance. The specific object is to detect the workpiece contour flange protrusion.The method comprises the steps that the transformation relation between workpiece coordinates and image coordinates is analyzed and inferred; an image region-of-interest is extracted under the guidance of a G-code contour in combination with the coordinate transformation relation, the processing method of mid-value filtering, image binarization and edge detection is performed on the extracted region-of-interest, and a workpiece contour edge is obtained; and a standard G-code contour is used as a benchmark to recognize and judge the edge protrusion, a minimum enclosing rectangle method is adopted to enclose the contour edge protrusion, and visual detection of the contour edge protrusion is realized. Through the machine vision detection method for the workpiece contour flange protrusion, starting point, ending point and height information of a feature segment of the workpiece contour flange protrusion can be effectively recognized.

The invention provides a network training method, an image processing method, a network, terminal equipment and a medium. The training method comprises the following steps: S1, acquiring a sample image accommodating a target object, a sample mask corresponding to the sample image and sample edge information corresponding to the sample mask; s2, inputting the sample image into an image segmentationnetwork to obtain a generated mask output by the image segmentation network; s3, inputting the generated mask into the trained edge neural network to obtain generated edge information output by the edge neural network; s4, determining a loss function according to the difference between the sample mask and the generated mask and the difference between the generated edge information and the sampleedge information; and S5, adjusting each parameter of the image segmentation network, and returning to the step S2 until the loss function is smaller than the threshold. According to the invention, the mask image output by the image segmentation network can represent the contour edge of the target object more accurately.

A method and system for quantifying profile characteristics of semiconductor devices, including receiving profile data for a device under evaluation and isolating from the profile data a region indicating a profile edge. The profile edge data is rotated by ninety degrees to become rotated profile edge data. The non-rotated profile edge data or rotated profile edge data is then used to calculate at least one geometric parameter describing the profile edge.

The invention provides a method for positioning a slight expanded target based on gesture compensation. The method comprises the following steps: firstly, pre-processing a to-be-processed image by adopting the Gauss smooth filtering to eliminate the influence of noise on a subsequent algorithm; secondly, obtaining a two-value edge of a single pixel by adopting an edge tracking method, and filtering peripheral points of the profile of a target to eliminate interfering points; thirdly, computing a momental ellipse of the target and parameters of the target by using the peripheral points of the profile of the target to confirm the gesture of the target; and finally, obtaining a translation parameter and a rotation parameter of the target by using the positioning method based on the gesture compensation. Therefore, through the adoption of the method, the fact that the target is positioned in the situations such as rotation, translation and scaling can be realized.

The invention relates to a real scene image cartoonalization processing method and device based on artificial intelligence, computer equipment and a storage medium. The method comprises the steps of obtaining a real scene image; based on the semantic information of the real scene image, performing image reconstruction processing and abstract smoothing processing on the real scene image to obtain an abstract cartoon image of the real scene image mapped on a cartoon domain; abstracting a cartoon image which has a significant structure in the real scene image and lacks contour edge lines in the real scene image; carrying out stylization processing on the abstract cartoon image to generate a style cartoon image with an artistic style; generating contour edge lines of the style cartoon image, and obtaining a cartoon image obtained after cartooning of the real scene image. By adopting the method, the quality of the generated cartoon image can be improved.

The invention discloses a tree three-dimensional visualization model realization method based on Sphere-Board. The method mainly solves the problems that an existing billboard or crossing plane model will deform when overlooked, the three-dimensional sense of the model is poor, and illumination calculation is hard to realize. According to the tree three-dimensional visualization model, a spherical curved surface is adopted to conduct geometrical approximation on an entire crown, and more details concerning branches and leaves are expressed through a texture synthesis method; given that outline edges of a projection of the crown are excessively smooth and continuous when the spherical curved surface is used for expressing the crown, the visual sense of reality of the edges of the crown will be weakened, so a shape remolding process is added for the rendered edges of the crown. The model is suitable for tree varieties with neat crown shapes and dense branches and leaves without holes. The method comprises the main steps of acquiring feature points of the crown to generate the spherical curved surface approximate to the geometrical shape of the crown, completing calculation of texture coordinates through the curved surface texture synthesis technology, dynamically extracting the rendered outline edges of the crown and remolding the shape of the outline edges.

The invention discloses an automatic cutting method. The automatic cutting method includes the steps that when a product to be cut is moved to a visual inspection area, the product to be cut is photographed by an industrial camera, and the image information of the product to be cut is obtained; the cutting patterns of the product to be cut are extracted by a vision algorithm program, and spline curve fitting is conducted on contour edge information of the cutting patterns to form smooth curve contour information; the product to be cut is moved to a cutting area, hand-eye calibration is carriedout by the relation between image pixel coordinates of the industrial camera and the tail end of a laser device, and smooth curve contour coordinates are converted to motion trail coordinates of thetail end of the laser device; and a vision industrial personal computer drives the laser device to operate along the motion trail according to the motion trail coordinates of the tail end of the laserdevice. The automatic cutting method has extremely high universality, images are automatically obtained by the industrial camera, the operation trail of the tail end of the laser device is obtained through coordinate transformation, the error caused by a tool and the product can be compensated during cutting, and the cutting precision is improved.

The invention provides an image matching and positioning method based on point feature and contour feature fusion. The method comprises the following steps of after Point feature detection, calculating segmentation thresholds, binarizing images, searching through contour edge points; performing contour approximate fitting to fuse point features and contour features, performing feature descriptionand feature matching, performing mapping matrix point screening, calculating a mapping matrix and a matching position, mapping a real-time graph center coordinate to a reference graph, and performingcompensation according to flight platform attitude information to obtain current geographic position information of the flight platform. Point features of the method have intrinsic contour constraints. Mismatching points can be quickly and effectively eliminated through contour features. The method has following advantages in that a mismatching problem caused by single point feature matching is avoided, matching correctness is improved, heterogeneous image matching positioning of the flight platform shooting real-time image and the satellite remote sensing image reference image is realized through point feature and contour feature fusion, and relatively high correctness is realized compared with a present matching algorithm.

The invention provides a steel bar size detection system and method based on image processing. The steel bar size detection system based on image processing comprises an image acquisition module whichis carried on an unmanned aerial vehicle and is configured to acquire a construction site picture; an image processing module which is configured to detect the edge of a steel bar in the constructionsite picture and extract an edge binary image; carrying out Hough straight line detection on the edge binary image, and fitting a steel bar contour edge; extracting the image containing the contour edge of the fitting steel bar again to remove shading impurities, then carrying out filtering and segmentation preprocessing, and finally detecting the contour size of the steel bar and the number of pixel points in each contour; obtaining the real area of the steel bar based on the real-time position height of the unmanned aerial vehicle, comparing the real area with the steel bar acceptance standard, and marking whether the steel bar is qualified.

A method and system for quantifying profile characteristics of semiconductor devices, including receiving profile data for a device under evaluation and isolating from the profile data a region indicating a profile edge. The profile edge data is rotated by ninety degrees to become rotated profile edge data. The non-rotated profile edge data or rotated profile edge data is then used to calculate at least one geometric parameter describing the profile edge.

Disclosed in the present application are an image processing method and device and a computer-readable storage medium. The image processing method comprises: acquiring grayscale values of a pluralityof pixels in an original image, wherein the grayscale value of a pixel corresponds to a certain reference value of a current grayscale division rule; performing sparsification processing on the current grayscale division rule so as to remove part of the reference values from the current grayscale division rule; adjusting the grayscale value of each pixel according to the current grayscale divisionrule after the sparsification processing to obtain a processed image; and classifying each pixel according to the adjusted grayscale value of each pixel so as to process the processed image by block;According to the described mode, the present application may well preserve clear contour edges of an object in an image and discard the small and fragmentary edges therein, which reduces the amount of calculation in image processing, and efficiently implements the image processing by block.

The invention discloses a shallow relief object recognition processing method based on an RGB monocular image. The method includes: reading RGB monocular image information, wherein the RGB monocular image information comprises the total number of pixels in the image, and the chromaticity information, the brightness information and the position information of each pixel; extracting a contour edge of the image by adopting an edge detail enhancement processing method to obtain contour edge information of an object in the image; segmenting the RGB monocular image by adopting an improved connecteddomain calibration algorithm on the basis of the contour edge to obtain an image region; solving the height value of each pixel in each image area through a pixel point brightness depth recovery algorithm, obtaining three-dimensional point cloud data of each image area, and constructing a shallow relief model through a triangular patch reconstruction algorithm. According to the method, the shallowrelief model can be well constructed and recognized through a common RGB image, the calculation resource consumption is low, the calculation amount is small, the efficiency is high, and a foundationis laid for tactile perception of the image.