Abstract

JPEG compression is an image compression technique that is used for a wide range of applications these days.\nThis paper describes a method to reduce blocking artifacts and quantisation noise through a machine learning approach.\nArtificial neural networks is used for this purpose. Feed –forward neural networks are used which make use of feedback and\nhence accuracy level is increased. Based on the number of hidden neurons and the quality of compression the regression\nvalue is obtained. Finally the DCT coefficients obtained are replaced by the actual pixel intensity values to obtain the final\noutput image. The regression value and least mean square error values are computed and the graph is plotted for the same.\nThe experimental results for the proposed method are also presented in this paper.