Improved Local Linear Embedding Algorithm And Its Application On Image Recognition

With the substantial increase of data, data processing technology is facing greatchallenges, especially the image data. In recent years, image data analysis technologybecomes a hotspot in research field. It is one of the important researches in the imagerecognition that extracted the inherent characteristics structure from the high-dimensional image data. Therefore, the key of image recognition is fully excavated theinherent characteristics structure of the image data. However, the traditional imagefeature extractions are mostly from the bottom such as color, texture, shape, etc. It isdifficult to extract the high-level information from the high-dimensional image data.Because the development of the methods of the manifold learning provides some waysfor the high-level information extraction of image, manifold learning and nonlineardimensionality reduction theory gains more and more attention.This paper is based on manifold learning, mainly studies and comparatives theLocally Linear Embedding (LLE) algorithm and Supervised Locally Linear Embedding(SLLE) algorithm. For the lack of two manifold learning algorithms, this paperconstructs a SSLLE manifold learning algorithm. The experiments show that SSLLEalgorithm can overcome the shortcomings of LLE algorithm and SLLE algorithm insome extent.The main work of this paper includes the following aspects:1. This paper studies on the methods and presents situation and applications fieldsof manifold learning, and detailed analysis the advantages and disadvantages of the LLE algorithm and SLLE algorithm. This paper summarized the theory of image recognitionand the methods of image extraction.2. To the poor noise immunity and difficult reconstruct the neighborhoodparameters of LLE algorithm and SLLE algorithm, this paper construct SSLLEmanifold learning algorithm. SSLLE algorithm by reconfiguration node makesneighborhood parameter automatically selected, and verified by experiment that itssensitivity to noise is lower than LLE algorithm and SLLE algorithm.3. This paper applies SSLLE manifold learning algorithm in image recognition,and achieved a better results of the image recognition by programming.