This paper presents a new approach to off-line handwritten numeral recognition based on structural and statistical features. Five different types of skeleton features: (horizontal, vertical crossings, end, branch, and cross points), number of contours in the image, Width-to-Height ratio, and distribution features are used for the recognition of numerals. We create two vectors Sample Feature Vector (SFV) is a vector which contains Structural and Statistical features of MNIST sample data base of handwritten numerals and Test Feature Vector (TFV) is a vector which contains Structural and Statistical features of MNIST test database of handwritten numerals. The performance of digit recognition system depends mainly on what kind of features are being used. The objective of this paper is to provide efficient and reliable techniques for recognition of handwritten numerals. A Euclidian minimum distance criterion is used to find minimum distances and k-nearest neighbor classifier is used to classify the numerals. MNIST database is used for both training and testing the system. A total 5000 numeral images are tested, and the overall accuracy is found to be 98.42%.

Texture refers to the variation of gray level tones in a local neighbourhood. The “local” texture information for a given pixel and its neighbourhood is characterized by the corresponding texture unit. Based on the concept of texture unit, this paper describes a new statistical approach to texture analysis, based on average of the both fuzzy left and right texture unit matrix. In this method the “local” texture information for a given pixel and its neighbourhood is characterized by the corresponding fuzzy texture unit. The proposed Average Fuzzy Left and Right Texture Unit (AFLRTU) matrices overcome the disadvantage of FTU by reducing the texture unit from 2020 to 79. The proposed scheme also overcomes the disadvantage of the left and right texture unit matrix (LRTM) by considering the texture unit numbers from all the 4 different LRTM’s instead of the minimum one as in the case of LRTM. The co-occurrence features extracted from the AFLRTU matrix provide complete texture information about an image, which is useful for texture classification. Classification performance is compared with the various fuzzy based texture classification methods. The results demonstrate that superior performance is achieved by the proposed method.