High dynamic range three-dimensional image registration method

Abstract

The present invention discloses a high dynamic range three-dimensional image registration method, and relates to the computer vision field. The improved SURF feature descriptor in the method combines an Haar-like feature to increase description of opposite side features and opposite angle features, in the matching process, the improved RANSAC algorithm selects relatively high-quality feature matching points as initial data to reduce the nondeterminacy of model parameters and improve the registration accuracy. The improved SURF algorithm solves the registration problem of multi-exposure image sequences in the condition that three-dimensional parallax cannot be neglected, has fast speed and is suitable for the real-time registration condition.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a high dynamic range three-dimensional image registration method. Background technique [0002] In recent years, high dynamic range image composition (High Dynamic Range Image Composition) technology has achieved high-tech development. Real-world scenes often have a high dynamic range, and traditional digital cameras struggle to capture the full dynamic range due to their limited bit depth. A compensatory measure is to take multiple images of the same scene with different exposures and combine them to indirectly restore the complete dynamic range. After using the tone mapping algorithm, the high dynamic image is compressed and displayed on the low dynamic range display components (such as LCD, CRT ) to improve image quality. [0003] In the basic process of digital camera imaging, the output of the camera is mainly determined by two factors. One is the brightness of the light in ...

AI Technical Summary

Problems solved by technology

SURF uses integral graph and template scaling to simulate scale changes, and the speed is fast. Its descriptor uses the statistics of the Haar wavelet response in the horizontal and vertical directions to construct features, which has a certain anti-noise ability, but as a statistical structure feature, the discriminative power of its descriptors is relatively low

[0005] In the matching of SURF feature points, the traditional RANSAC algorithm has a stable process, strong robustness to matching error feature points and noise, and good ability to eliminate wrong matches. However, the RANSAC algorithm estimates parameters through continuous iteration and The test is completed, and the initial model parameters are calculated from randomly selected data, and the uncertainty is relatively large; if the error of the randomly selected initial data is large, the performance of the algorithm will be seriously affected

Images