Based On The Manifold Learning For Underwater Image Dimension Reduction Research

Ocean owns great significance for China’s sustainable development strategy, andoceanographic information is the first condition of the development of marine space.Underwater images is the main way to get access to information, and due to the morecomplex environment of the deep underwater imaging, the process of acquisition,storage, and transmission in the underwater imaging system have been influenced by avariety of factors. So that the details of ultimate underwater color of the imageappears in the visual are fuzzy, causing the phenomena such as decreasing in contrast.And the reason of large amount of data transmission leads to timeliness. How to getthe underwater color images fast and efficiently with more important information hasbecome an important research topic.Usually underwater color image is distorted, showing a single blue atomizationphenomenon. Results not only fail to achieve the effect of general color images inbright and contrast almost compared to black and white images. As multi-channelcolor superimposed, the detail texture partially obscured. In this exceeds thethree-dimensional images of human perception, the human eye is difficult to identify,Dimension reduction technology is an important means to solve it. Manifold learningmethod for processing image, the three-dimensional space data is mapped to atwo-dimensional, and as much as possible to ensure that the data geometricrelationships and the same distance measure. Through the process of dimensionalityreduction, the main features contained in the underwater image can dig out moreuseful information, at the same time lowering the small amount of data in thedimension, to facilitate real-time transmission.The flow of this study is shaped to learn algorithm underwater color imageprocessing. On the basis of imaging principle of underwater color images andenhanced recovery, and through the process of three nonlinear dimensionalityreduction algorithm theory of learning, we can reduce the dimension of underwater color images from both algorithms ideological differences, contrast, and drop thedimensional effect evaluation parameters according to comparison. Proved byexperiments, manifold learning has better results in dimensionality reduction ofunderwater color images, and the internal structure of the information obtained fromthe results has more details which are clearer than others. And it also compresses theamount of data better with timeliness enhanced.