CN-121981907-A - Image synthesis method, device, computer equipment and storage medium

CN121981907ACN 121981907 ACN121981907 ACN 121981907ACN-121981907-A

Abstract

The application discloses an image synthesis method, a device, computer equipment and a storage medium, which are characterized in that key technologies such as denoising filtering, feature extraction, matching screening based on a distance ratio, geometric transformation and the like are introduced into a multi-source image processing link, and then the key technologies such as edge fusion are combined, specifically, denoising treatment effectively reduces the influence of imaging noise on feature extraction and descriptor stability, robust feature point matching and abnormal elimination ensure high precision of registration under a cross-view condition, splicing dislocation caused by mismatch is avoided, the accuracy of a corrected image on space mapping is ensured based on transformation matrix estimation of a refined matching point pair, and the common splicing problems such as image edge brightness difference, texture fault and the like are further solved by an edge fusion technology, so that overlapping area transition is natural and has no obvious gap. According to the scheme, high-precision automatic splicing of multi-view and multi-source images can be realized, and the space alignment quality and visual rendering effect of the synthesized images are remarkably improved.

Inventors

ZHOU CHANGLIN
Yin Jiehao
CUI JIAN
LIU JIE

Assignees

彩迅工业(中山)有限公司

Dates

Publication Date: 20260505
Application Date: 20251223

Claims (10)

1. An image synthesizing method, characterized by comprising: acquiring a multi-source image, and carrying out denoising and filtering treatment on the multi-source image to obtain a smooth image; Extracting key feature points of the multi-source image by adopting a target detection algorithm to form a feature point set, wherein the feature point set at least comprises feature point positions; Calculating a neighbor distance ratio between two adjacent images aiming at the characteristic point set, and determining an initial matching point pair set according to the neighbor distance ratio, wherein the matching point pair set reflects a cross-view angle corresponding relation; based on the initial matching point pair set, adopting an abnormal elimination algorithm to eliminate abnormal matching characteristic points to obtain a refined matching point pair set; Calculating an image geometric transformation matrix according to the refined matching point set pair to obtain a transformation matrix, wherein the transformation matrix represents deformation of an image; Applying an image correction transformation to the smoothed image based on the transformation matrix to obtain a corrected image; and merging the edge areas of the corrected images based on the refined matching point pair sets to obtain a composite image.
2. The image synthesis method according to claim 1, wherein the step of calculating a neighbor distance ratio between two adjacent images for the feature point set, and determining an initial matching point pair set according to the neighbor distance ratio, specifically comprises: Sequentially calculating Euclidean distances between the characteristic points of the two images according to the positions of the characteristic points in the first image and the positions of the characteristic points in the second image to obtain a distance set, wherein the first image and the second image are two adjacent images; Sequentially selecting a first neighbor distance and a second neighbor distance from the distance set; Calculating a distance ratio between the first neighbor distance and the second neighbor distance to obtain a neighbor distance ratio; Comparing the neighbor distance ratio with a preset ratio threshold value, and determining whether two feature points corresponding to the neighbor distance ratio form candidate matching points or not; and adding the characteristic point pairs forming the candidate matching points into the initial matching point pair set.
3. The image synthesis method according to claim 2, wherein the step of comparing the neighboring distance ratio with a preset ratio threshold value to determine whether two feature points corresponding to the neighboring distance ratio constitute candidate matching points, specifically includes: Judging whether the neighbor distance ratio is smaller than or equal to the preset ratio threshold; If the neighbor distance ratio is smaller than or equal to the preset ratio threshold, marking two feature points corresponding to the neighbor distance ratio as candidate matching points; And if the neighbor distance ratio is larger than the preset ratio threshold, marking the two feature points corresponding to the neighbor distance ratio as non-candidate matching points.
4. The image synthesis method according to claim 1, wherein the step of calculating an image geometric transformation matrix according to the refined matching point pair set to obtain a transformation matrix specifically comprises: establishing a linear constraint equation of an image geometric transformation model based on the refined matching point pair set; solving the linear constraint equation by adopting a least square method to obtain initial geometric transformation parameters; Performing error evaluation on the initial geometric transformation parameters, and adjusting or performing iterative optimization on abnormal parameters in the image geometric transformation model according to error evaluation results; and constructing an image geometric transformation matrix according to the optimized geometric transformation parameters to obtain the transformation matrix.
5. The image synthesizing method as set forth in claim 1, wherein the step of applying an image correction transform to the smoothed image based on the transform matrix to obtain a corrected image includes: determining the mapping position of each pixel in the smooth image in a target coordinate system according to the transformation matrix; carrying out space resampling on the smooth image based on the mapping position to obtain a resampled pixel value; performing pixel interpolation calculation on the smooth image based on the resampled pixel value to obtain pixel interpolation data of the smooth image; and carrying out pixel integration on the original pixel data of the smooth image and the pixel interpolation data to generate a corrected image.
6. The image synthesis method according to claim 5, wherein the step of performing pixel interpolation calculation on the smoothed image based on the resampled pixel values to obtain pixel interpolation data of the smoothed image specifically includes: acquiring resampling pixel positions needing interpolation in the smooth image; Selecting a plurality of original pixel points in the neighborhood range of the resampling pixel position as interpolation reference pixels; Calculating a corresponding interpolation weight according to the spatial distribution relation of the interpolation reference pixels; carrying out weighted summation on the pixel value of the interpolation reference pixel based on the interpolation weight to generate a target interpolation pixel value; and recording the target interpolation pixel value as pixel interpolation data of the smooth image.
7. The image synthesis method according to claim 1, wherein the step of merging edge regions of the corrected image based on the refined matching point pair set to obtain a synthesized image specifically comprises: determining an overlapped edge area between two adjacent corrected images according to the refined matching point pair set; Performing corresponding relation analysis on pixels in the overlapped edge area, and determining fusion weights of all pixels in the overlapped edge area; weighting and fusing the corresponding edge pixels of the two corrected images based on the fusion weight; Performing brightness consistency adjustment on the fused overlapped edge areas; and splicing the overlapped edge area subjected to fusion and brightness adjustment with the non-overlapped area of the corrected image to generate the synthetic image.
8. An image synthesizing apparatus, comprising: The denoising and filtering module is used for acquiring a multi-source image, and denoising and filtering the multi-source image to obtain a smooth image; The characteristic point extraction module is used for extracting key characteristic points of the multi-source image by adopting a target detection algorithm to form a characteristic point set, wherein the characteristic point set at least comprises characteristic point positions; The distance ratio module is used for calculating the adjacent distance ratio between two adjacent images aiming at the characteristic point set, and determining an initial matching point pair set according to the adjacent distance ratio, wherein the matching point pair set reflects a cross-view angle corresponding relation; the characteristic point eliminating module is used for eliminating abnormal matched characteristic points by adopting an abnormal eliminating algorithm based on the initial matching point pair set to obtain a refined matching point pair set; The geometric transformation module is used for calculating an image geometric transformation matrix according to the refined matching point set pair to obtain a transformation matrix, wherein the transformation matrix represents the deformation of the image; the image correction module is used for applying image correction transformation to the smooth image based on the transformation matrix to obtain a corrected image; And the image synthesis module is used for merging the edge areas of the corrected images based on the refined matching point pair set to obtain a synthesized image.
9. A computer device comprising a memory and a processor, the memory having stored therein computer readable instructions which when executed by the processor implement the steps of the image compositing method of any of claims 1-7.
10. A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor implement the steps of the image synthesis method according to any of claims 1 to 7.

Description

Image synthesis method, device, computer equipment and storage medium Technical Field The application belongs to the technical field of image processing, and particularly relates to an image synthesis method, an image synthesis device, computer equipment and a storage medium. Background In the field of modern vision technology, image synthesis technology is very important due to its wide application in monitoring, virtual reality, panoramic imaging and other scenes. The technology can integrate the multi-source images into a complete picture, and provides more comprehensive visual information for users. However, despite its importance, it is self-evident that current research and applications still face many challenges, with breakthrough being needed to meet the increasing demands. When the existing method is used for realizing image synthesis, the coordination of the multi-view images and the naturalness of the whole picture are often difficult to balance. Many schemes are prone to problems of unnatural picture transitions or information loss when processing images acquired at different angles, especially in complex environments where the variability between images can further amplify the defect. This limitation not only affects the quality of the final picture, but also limits the application of the technology in demanding scenarios. Focusing on the technical difficulties, one of the core problems in image synthesis is how to handle image linking at multiple camera angles. The images shot by different cameras often cause geometric deformation of the edges of the images due to the angle and position difference, and the deformation can cause obvious dislocation or discontinuous phenomenon of the images during splicing. More importantly, such geometric deformation can further cause difficulty in matching feature points, as key points in an image may exhibit different morphologies or positional offsets due to changes in viewing angle, making it difficult for the system to accurately identify and align the points. Taking a specific scene as an example, in the production and processing process, pictures of different areas of a processed workpiece are respectively shot through a plurality of cameras, if one part at the edge of a picture appears to be inclined in a transverse camera and appears to be vertical in a longitudinal camera, the form difference of the parts can cause that the system can not correctly judge that the parts are the same object during splicing, and finally obvious fracture or overlapping of the spliced part of the pictures is caused. Therefore, how to effectively solve the contradiction between geometric deformation and feature point matching under the multi-camera view angle, and ensure the naturalness and accuracy of picture splicing becomes a key problem to be overcome in the research. Disclosure of Invention The embodiment of the application aims to provide an image synthesis method, an image synthesis device, computer equipment and a storage medium, so as to solve the contradiction between geometric deformation and characteristic point matching under the multi-camera view angle and ensure the naturalness and accuracy of picture splicing. In order to solve the above technical problems, an embodiment of the present application provides an image synthesis method, which adopts the following technical scheme: an image synthesis method, comprising: acquiring a multi-source image, and carrying out denoising and filtering treatment on the multi-source image to obtain a smooth image; extracting key feature points of the multi-source image by adopting a target detection algorithm to form a feature point set, wherein the feature point set at least comprises feature point positions; Aiming at the characteristic point set, calculating the adjacent distance ratio between two adjacent images, and determining an initial matching point pair set according to the adjacent distance ratio, wherein the matching point pair set reflects the cross-view angle corresponding relation; based on the initial matching point pair set, adopting an abnormal elimination algorithm to eliminate abnormal matching characteristic points to obtain a refined matching point pair set; Calculating an image geometric transformation matrix according to the refined matching point set pairs to obtain a transformation matrix, wherein the transformation matrix represents the deformation of the image; applying image correction transformation to the smooth image based on the transformation matrix to obtain a corrected image; and merging the edge areas of the corrected images based on the refined matching point pair sets to obtain a composite image. In order to solve the above technical problems, the embodiment of the present application further provides an image synthesis apparatus, which adopts the following technical scheme: an image synthesizing apparatus comprising: the denoising and filtering module is used for acquiring a multi-sour