Image Segmentation Using Artificial Neural Network and Genetic Algorithm: A Comparative Analysis

Image segmentation is an important step in image processing. Most of the segmentation methods are parametric and the results of segmentation depend on the correctness of the estimated parameters. In case of supervised segmentation, a priori knowledge is needed for successful segmentation. So, nonparametric and unsupervised segmentation method is used when a priori information is not available. Kohonen's Self Organizing Maps (SOM), an unsupervised and nonparametric artificial neural network method is used to identify the main features present in the image. Genetic Algorithm (GA) can be applied to the results of SOM for optimal segmentation results. In this paper, the basic SOM, SOM combined with GA and some of the variants of SOM like the Variable Structure SOM (VSSOM), Parameterless SOM (PLSOM) are compared and their performance is evaluated. A new unsupervised, nonparametric method is developed by combining the advantages of VSSOM and PLSOM. The experiments performed on the satellite image shows that the modified PLSOM is efficient and the time taken for the segmentation is less when compared to the other methods.