Semantic Perceptual Image Compression using Deep Convolution Networks

TL;DR

This paper introduces a deep learning-based method for semantic image compression that enhances visual quality by identifying and encoding salient regions at higher quality levels, compatible with standard JPEG decoders.

Contribution

It presents a novel CNN architecture that detects semantic regions without object labeling, enabling content-aware compression with minimal complexity increase.

Findings

Achieves higher visual quality at the same compression size.

Compatible with standard JPEG decoders.

Effective on Kodak and MIT datasets.

Abstract

It has long been considered a significant problem to improve the visual quality of lossy image and video compression. Recent advances in computing power together with the availability of large training data sets has increased interest in the application of deep learning cnns to address image recognition and image processing tasks. Here, we present a powerful cnn tailored to the specific task of semantic image understanding to achieve higher visual quality in lossy compression. A modest increase in complexity is incorporated to the encoder which allows a standard, off-the-shelf jpeg decoder to be used. While jpeg encoding may be optimized for generic images, the process is ultimately unaware of the specific content of the image to be compressed. Our technique makes jpeg content-aware by designing and training a model to identify multiple semantic regions in a given image. Unlike object detection techniques, our model does not require labeling of object positions and is able to identify objects in a single pass. We present a new cnn architecture directed specifically to image compression, which generates a map that highlights semantically-salient regions so that they can be encoded at higher quality as compared to background regions. By adding a complete set of features for every class, and then taking a threshold over the sum of all feature activations, we generate a map that highlights semantically-salient regions so that they can be encoded at a better quality compared to background regions. Experiments are presented on the Kodak PhotoCD dataset and the MIT Saliency Benchmark dataset, in which our algorithm achieves higher visual quality for the same compressed size.