Recognition of Contour Invariants with NeuroFuzzy Classifier

In this study, we explore contour invariants for handwritten digits recognitions with neuro-fuzzy classifier. We use fuzzy triangular function in backpropagation network to initialize the weights. The results reveal that fuzzy triangular membership function manages to decrease the network convergence rate with proper parameter setting. In this study, unthinned images are appropriate for training and classification purpose as it preserves the images significant features. From our experiments, the results show that contour invariants exhibits highest rate of classification compares to geometric and zernike invariants.