Self-organising map for shape based image classification

The main questions of image classification ape how to identify images and how they are to be classified. The proposed system in this paper uses run lengths for classification of the images into groups, because we assume that run length approximates the shape and texture information of an image. We have selected an unsupervised neural network called a Self-Organising Map to solve the classification problem. The run length histograms are computed, normalised and insignificant features are discarded. The resulting image histograms are stored in a database. The training of the neural network and classification are accomplished using the run length histograms obtained from the image histogram database. The performance of the system was tested by comparing the classifications achieved by the neural network with object-based classifications, which were intuitively determined.