Modified Image Compression by Combined Neural Network and Arithmetic Coding Processing

Modified Image Compression by Combined Neural Network and Arithmetic Coding Processing

This article emphases the skeleton for image compression-decompression by using (Perceptron Neural Network and Arithmetic Coding). At first compression of each three-pixel into single value, this value is called Compression value. In our work the neural network is not needed for weight training at compression part, instead, the weights used in our work are one dimensional array containing floating point values. The total of weights values equal one. In the decompression part, the weights become an input to neural networks. Later, neural networks update the pixels according to error between Compression value and desired output to become approximately original three-Bytes. The second stage is Arithmetic coding algorithm uses to convert vector of compression values in to single floating point number. Our approach tested with three types of images and different size; also in this article the performance of the algorithm is computed.