An Unsupervised Segmentation Method for Dynamic Scenes with Complex Objects Based on Convex Optimization of Multi-Scale Combined Features
A dynamic scene and combined feature technology, which is applied in the fields of image processing and computer vision, can solve the problems of inconsistent appearance, color, shape topology, change over time, difficult foreground target effective segmentation, etc.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0058] Embodiment 1: as Figure 1-8 As shown, an unsupervised segmentation method for complex target dynamic scenes based on multi-scale combined feature convex optimization, first extracts high-scale target features in the wavelet domain, calculates motion and spatial edge responses, and fuses wavelet domain, spatial and temporal edge features Obtain the multi-scale combined probability edge response, and then establish the weighted total variational energy functional model of the internal mapping of the combined edge, use the alternating direction to calculate the weighted total variational energy functional model convex optimization, and define the superpixel scale space-time Markov random field The unary energy function item and the binary space energy function, and finally obtain the posterior probability segmentation result of the image sequence through one-step iterative reasoning of graph segmentation;
[0059] The specific steps of the complex target dynamic scene uns...
Embodiment 2
[0096] Embodiment 2: as Figure 1-8 As shown, an unsupervised segmentation method for complex target dynamic scenes based on multi-scale combined feature convex optimization, input image sequence data containing target information and dynamic scene information of the target, calculate multi-scale wavelet edge features, structured random forest space Edge features and motion gradient features, through the maximum edge response of the foreground target contour in the dynamic image sequence, the combined probability edge of the target contour feature is extracted; the total variation convex optimization model of the foreground position mapping seed point is constructed on a fine scale, and the variable is updated alternately. Smooth the target contour features extracted on a high scale, combine the probability edge internal mapping seed points, and establish the univariate data energy function and binary space and time energy function items of the Markov random field according to ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More