Research On Microscopic Image Identification Of Marine Phytoplankton Based On Fractal

Classification of marine phytoplankton is an important aspect for the study of marine ecology environment and the investigation of marine. The subject of this paper comes from the National Natural Science Foundation of China"The cell image identification of marine phytoplankton in China's coastal waters"(60572064). Based on fractal theory, this paper studies the identification of the microscopic image of marine phytoplankton by acquiring image from multiviewpoints with different growing periods and different angle.Fractal theory is a very active mathematics branch in modern non-linear, and is widely used in image processing. Especially, following the development of information technology, fractal has become one of the hot spots in the area of non-linear science. Fractal coding has been rapid developed because of its potential of high compression, and in recent years it has been used in image identification area.The fractal neighbor distance gives a quantitative measure of the input–output characteristics of the fractal code of an encoded image. As fractal coding carries the distribution characteristic of the image spatial domain, and the attractor of fractal code has uniqueness, it can be used for image matching, and thus used for image identification.Based on fractal theory and fractal coding, this paper does in-depth study on the matching method of fractal neighbor distance, and makes some improvement to the traditional method for the cell image of marine phytoplankton. Proved by the experiments, improved method is better than the traditional method. This paper mainly contains the following work:1. Fractal neighbor distance (FND) reflects the similarity between unknown image and decoding image with once iteration. But in case of once iteration, the decoding images can't convergence and have blocking effect. For the microscopic image of marine phytoplankton, this paper makes certain improvement on the matching method of FND. Under the premise of not affect the decoding efficiency, this paper changes iteration to four times, which makes the decoding image convergence and eliminate the blocking effect. Compared with the traditional FND, this paper proves the validity and reliability of the improved FND by image matching experiments of cell image of marine phytoplankton.2. For the cell image of marine phytoplankton, the identification method of improved FND also has limitations. During the process of algal image taken, the acquired images will have the difference in illumination and angel, because of the shooting condition and object. For different images with large differences in illumination and angel, it will cause error by using the matching method of FND. As singular value decomposition (SVD) has invariance in rotation and illumination, incorporating FND with SVD will reduce this error. Experiments show that, compared with FND fractal singular value neighbor distance can improve identification rate.