Abstract The problem of searching a similar pattern is an exciting and challenging research field of pattern recognition. The intelligence of humans for vision to read is a crucial phenomenon for machine simulation and has been carried out for a few decades. Therefore, in this article, a recognition system of handwritten Meitei Mayek (Manipuri script) is introduced using a convolutional neural network. Generally, character recognition is performed using the gray scale of the image of characters. However, we have additionally considered the corresponding gradient direction and gradient magnitude images to create three‐channels image for every character so that supplementary information from gradient images can be obtained for efficient recognition. Experiments are conducted on 14 700 sample images collected from various individuals of different age groups and educational backgrounds. A recognition rate of 98.70% is obtained, which is compared with the existing methods, and it is found to be superior performance than other neural network methods on Meitei Mayek.

It is almost impossible to distinguish between the original and manipulated images subjectively due to the growth in technologies. The main aim of this paper is to develop a robust hash function, which can withstand legitimate modifications. A robust hash algorithm based on scale-invariant feature transform (SIFT) and singular value decomposition (SVD) is being proposed in this paper. Hash values are generated from the maximum singular values obtained after applying SVDs to the SIFT key-points mapped non-overlapping blocks of pixels. The experimental result shows that the proposed hashing method is robust against the various attacks.