Image processing method and device, storage medium, and image device
An image processing and image technology, applied in the image field, can solve the problems of poor image texture quality, target tracking failure, low accuracy, etc., to reduce the large surface deformation error, improve the matching accuracy, improve the accuracy and the success of tracking rate effect
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
example 1
[0221] This example provides an image processing method that can be used for object tracking, including:
[0222] Obtaining a first feature point set of a reference image, where the reference image may be the first frame image or the previous frame image, which may correspond to the aforementioned first image;
[0223] Obtain the second feature point set of the currently input current frame image, where the current frame image is the aforementioned second image;
[0224] Using the surface deformation finally obtained by optimizing the previous frame image as the initial candidate surface deformation of the image frame, calculating the point pair matching between the first feature point set and the second feature point set;
[0225] Based on the point pair matching, the projection error of the point pair matching is obtained, and the deformation of the candidate surface is reconstructed; the projection error here can be the matching error between the aforementioned two matching...
example 2
[0230] Restoring the shape of the object with a non-rigid surface in the second image can involve three steps, as follows:
[0231] (1) Feature point correspondence: use the local texture information calculated from the feature point descriptor algorithm to establish a feature point matching relationship;
[0232] (2) Outlier rejection: Eliminate incorrect matching relationships by measuring their geometric compatibility with deformable models;
[0233] (3) Shape reconstruction, where the shape reconstruction is equivalent to obtaining a surface deformation: the non-rigid shape of the target surface is estimated based on the known template and the established feature point matching relationship.
[0234] Feature point correspondence refers to extracting feature points from a given image and then associating feature points with feature points in a nearest-neighbor manner through a suitable distance metric. When detecting feature points, feature point detectors and descriptors ...
example 3
[0285] This example provides an image processing method based on Example 1 and / or Example 2, including: graph construction, candidate matching filtering and adaptive outlier rejection.
[0286] Graph construction:
[0287] An undirected graph with n nodes can be represented as in represent the set of points and the set of edges, respectively. Given the initial region of the target surface of interest in the reference image Create a model graph for the surface as follows. A node here can be regarded as a feature point in the image.
[0288] Node Generation: Typically, feature points are extracted from images to represent local regions, and then they are modeled as vertices of the graph. Many feature-based methods obtain feature points as local minima / maximum values of DoG images across scales, that is, SIFT features. However, the number of feature points obtained by this method cannot be controlled, so the number of feature points obtained depends on the operator a...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More