Research Of Improved Algorithm Based On KNN And Its Application In Image Denoising

Preprocessing of data is an important task in data mining. In the era of data explosion, it is necessary to find valuable information in the massive data no matter what field we are in, and the pretreatment has become one of the necessary links. As one of the most important branch of data analysis, classification algorithm logically becomes an important member of the big family of pretreatment.KNN algorithm is an inert classification algorithm which could describe rules without a priori statistical knowledge and additional training data. Besides, it is very easy to be implemented. KNN algorithm is one of the classical algorithms of data mining. It is widely applied in many fields, such as the classification, regression, missing value fill and machine learning. However, it is inevitable that the algorithm has many problems, such as how to determine the appropriate values of K, the data processing effect of some special distribution is not ideal, and the computation complexity is not acceptable when the dimension of the data is too high. In order to solve these problems, researchers put forward many corresponding improved algorithms.Aimed at improving the neighbor selection strategy in KNN method, this paper expounds an improved KNN algorithm combined with local linear constraint coding. This method improved the classic algorithms which is sensitive to the data distribution with using traditional distance measurement. On the other hand, the combination of KNN method and image denoising will be introduced into the image processing field, in order to show the application prospect of KNN in image processing.The main work of this paper includes:(1) According to the nearest neighbor selection strategy of KNN algorithm, an improved algorithm is proposed. Combined with the theory of sparse coding and local constraint linear coding, the KNNLC algorithm is proposed to improve the classical KNN algorithm. Experiments show that the average effect of KNNLC algorithm in classification performance is better than the classical KNN algorithm.(2) For solving the problems of classical mean filter which may cause the picture fuzzy, this paper proposed the improvement method for the average filtering algorithm by using KNN principle and fuzzy set theory. By filtering every pixel of the image with noise, the experiments showed that using improved filtering algorithm to preprocess is more stable and more excellent than using classical mean filter and median filtering algorithm which is widely recognized.Through the analysis of theory and experiment, this paper illustrates the application prospect of the method which is based on the idea of K nearest neighbor in data preprocessing especially in image data preprocessing field.