Improving Image Classification using Fuzzy Neural Network and Backtracking Algorithm

Abstract

We propose an improved image classification method using fuzzy neural network (EFNN) which describes an algorithm in order to create a class rule based on the training data image. Redundant rule on 2 or more of data training image is generated in some data processing. A solution to the problem using backtracking algorithm, which will determine the appropriate class rule, is used by one of the training data images. Thus, every rule has an image of the appropriate class. In the process of inputting the data EFNN algorithm, 7 statistical parameters are used as a representation of the image characteristics, for feature extraction using wavelet Haar 2. The image becomes more leverage and has different characteristics to the representation of the image of the other. All input from crisp number is converted into fuzzy number with 5 membership function, which are Very Low, Low, Medium, High and Very High. Here, each image is represented by 7 statistical parameters and each parameter is divided into 5 categories. Percentage of accuracy in the classification process by using algorithms EFNN is above 95 percent for all data training, especially when it is compared with the original FNN.