Objective Evaluation of Image Segmentation Algorithms Using Neural Network
کلمات کلیدی:
objective evaluation; neural network; imagesegmentation; important feature ; over-segmentation
چکیده:
Image segmentation is an important research area incomputer vision and many image segmentation methods havebeen proposed, therefore it is necessary to be able to evaluate theperformance of image segmentation algorithms objectively. Inthis paper we present a new metric to evaluate the accuracy ofimage segmentation algorithms, based on the most importantfeature of each segments using neural networks. The neuralnetwork after training can assess the similarity or dissimilarity ofeach pairs of segments, based on the most important feature oftwo segments that can be distinguished from each other andfinally the segmentation algorithms accuracy have beencomputed by novel presented metric. Our proposed method donot require a manually-segmented reference image forcomparison, therefore can be used for real-time evaluation and issensitive to over-segmentation. Experimental results wereobtained for a selection of images from Berkeley segmentationdata set and demonstrated that it’s a proper measure forcomparing image segmentation algorithms.