Research On Feature Extraction And Matching For Underwater Unstructured Environment

Nowadays, both image technology and image processing system have comprehensive applications in the areas like ocean engineering, aviation industry, biomedicine, communication engineering, military security and agriculture. Image processing technologies take great importance especially upon ocean surveillance, seabed prospect and underwater target detection. Moreover, the hardware and software system that support these technologies are AUVs and image processing system. In the image processing system, the complexity of environment will lead to much uncertain factors that are dynamic, unstructured and edge blurring. These factors may be of impediments to targets detection and recognition, especial those targets that are unstructured. So of course, this is one of the hot spot of current image processing area.This paper mainly investigates feature extraction and matching methods in unstructured environment. We first analyzed some nowadays investigations in the area of interest, briefly summarized the feature extraction and matching methods and relative technology fields. Then, the Scale Invariant Feature Transforms (SIFT) is introduced as a best feature extraction method of the area. Based on the comprehension we got from the arithmetic, matlab platform is constructed for experiments. Meanwhile, some improvements are made on the matching process, such that the discussed Gaussian multi-scale invariant feature extraction method can be adapt to underwater unstructured environment. In the experiments, we evaluate the performance of method upon several aspects, i.e. image rotation, scale variation, illumination change, transmission projection, image blurring and noise adding. Also, the method is applied to real natural scenes taken from underwater unstructured environment. The experimental results shows that, such Gaussian multi-scale invariant feature extraction method is invariant to image rotation, translation and scaling, not sensitive to illumination change and projection change, can be adapt to underwater unstructured environment. This method has significant investigation value in the fields of environmental machine perception, map construction and target localization. Lastly, in order to extend its application range, color feature extraction is analyzed. The potential value of the method in the color image is presented, which can provide reference to the future work.