Efficient Image Compression by Machine Learning

Media broadcasting with video streaming is booming day by day, image data compression is becoming essential. The key goal of image compression is to obtain a relatively small bit rate and a good visual quality of decompressed images. Artificial intelligence and machine learning are such technologies to carry out this task. Each picture is composed up of pixels. Those pictures are labeled as noisy pictures. We propose a convolution auto encoder neural network to compact the images by using the MNIST database where we up sample and down sample an image to increase and decrease its quality. We take 128 by 128 dimensional image. By generating a deep learning machine image, the image must be compressed and transformed to 128 by 1 dimensional vectors in order to obtain a clear original picture type. The main objective is to compress the picture without reducing its quality, predicting the value present in the compressed picture of the MNIST database.