International Journal of Computer Theory and Engineering

Abstract

The volume of unsolicited commercial mails has grown extremely in the past few years because of increased internet users. This unsolicited mails termed as spam occupy large storage space and bandwidth. Therefore designing an efficient spam filter is a challenging issue ahead for the future generation. Here we use gradient histogram is a key feature that can be exploited to improve the categorization capability. The gradient values are valuated for each pixel of an image. These obtained features are then normalized for efficient spam classification. These normalized features are then applied as input for feed forward back propagation neural network (BPNN) model. The experiments are conducted for different training/testing rule for the BPNN. The performance measure in terms of accuracy is determined for the proposed method using various training rule of BPNN.