DCT-CompCNN: A Novel Image Classification Network Using JPEG Compressed DCT Coefficients

Abstract

The popularity of Convolutional Neural Network (CNN) in the field of Image Processing and Computer Vision has motivated researchers and industrialist experts across the globe to solve different challenges with high accuracy. The simplest way to train a CNN classifier is to directly feed the original RGB pixels images into the network. However, if we intend to classify images directly with its compressed data, the same approach may not work better, like in case of JPEG compressed images.This research paper investigates the issues of modifying the input representation of the JPEG compressed data, and then feeding into the CNN. The architecture is termed as DCT-CompCNN. This novel approach has shown that CNNs can also be trained with JPEG compressed DCT coefficients, and subsequently can produce a better performance in comparison with the conventional CNN approach.The efficiency of the modified input representation is tested with the existing ResNet-50 architecture and the proposed DCT-CompCNN architecture on a public image classification datasets like Dog Vs Cat and CIFAR-10 datasets, reporting a better performance.

An Introduction to Neural Data Compression

“…Training modern classifiers requires the decoding of millions of images or video frames. For example, instead of decoding JPEG images to pixels, Bulla et al [163] only decoded the DCT coefficients and fed them directly to a classifier.…”

DCT-Former: Efficient Self-Attention with Discrete Cosine Transform

“…Previously [44] proposed to operate a Convolutional Neural Network (CNN) on the DCT coefficients of a JPEG compressed image to avoid the need to run the full JPEG decoding algorithm. Several other contributions, such as [45][46][47][48], explored similar concepts for computer vision problems with varying degrees of success.…”

Video Quality Enhancement Using Deep Learning-Based Prediction Models for Quantized DCT Coefficients in MPEG I-frames

“…Rajesh et. al [7] propose the DCT-CompCNN, a CNN that uses DCT coefficients as the input for a classification task. They tested this modified input representation by comparing the existing ResNet-50 [12] architecture and their proposed architecture using a public dataset, reporting a better performance.…”

“…Many recent works have successfully applied DL architectures, specially some types of Convolutional Neural Networks (CNNs), to reduce the noticeable distortion resulting from the lossy JPEG/MPEG compression technique. Many of them are built upon the processing made entirely on the spatial domain, usually based on learned filters applied directly over the image pixels or blocks [3], [4], [5], [6], while others opt to make adjustments (and learning) in the frequency domain (usually combined with spatial information) based on the Discrete Cosine Transform (DCT) which is part of the compression process [7], [8], [9], [10].…”