Abstract:With the rapid development of the Internet technology and the digital imaging device,massive pictures are uploaded and saved in the network disks every day.Traditional image coding methods (such as JPEG,JPEG2000) only exploit spatial correlations inside a picture,thus their coding efficiency cannot satisfy the demand of the huge picture data.In order to compress the images in the Internet more efficiently,we proposed a novel cloud-based image compression method by taking advantage of the ‘ big data’ challenge,rather than simply treating it as a difficulty.The proposed system adopts the processing mode of ‘encoding one image immediately after it is uploaded’,which effectively adapts to the rapid change of the big-data environment.It compresses an image by fully exploiting the correlations between the image-to-be-compressed and other images in the cloud.We had three innovative technical designs to serve for the compression framework.(a) We designed a scheme based on an efficient image retrieval technique to effectively filter the ‘big data’ and to only keep the ‘small data’ that well matches the image to be compressed.(b) By utilizing the selected ‘small data’,we did inter-image prediction coding,which is essential for the performance of the compression system.We exploited the techniques of block matching-based prediction coding and rate-distortion optimization,so that coding bits are tremendously reduced in the premise of good reconstruction quality.(c) We ensured the prediction accuracy by applying a couple of preprocessing techniques on the retrieved similar image to obtain good references.In this step,we applied projective transformation to align the retrieved image and the current image,and did illumination compensation to make the values of corresponding pixels as close as possible.This paper presents a novel framework for image compression in the big-data era,which tremendously saves storage requirement for images in the cloud disks.The proposed system has significant compression performance gains compared to traditional image coding methods and even the advanced HEVC intra coding.The experimental results demonstrate that it outperforms JPEG and HEVC intra coding by 78.5％ and 67.2％ on average,respectively.Moreover,compared to the existing state-of-the-art attempt for cloud-based image coding in the literature,both the objective and subjective reconstruction quality of our algorithm have obvious improvement.

Efficient Feature Compression for Edge-Cloud Systems

Flexible Variable-Rate Image Feature Compression for Edge-Cloud Systems

CDC: Classification Driven Compression for Bandwidth Efficient Edge-Cloud Collaborative Deep Learning

Feature Compression for Cloud-Edge Multimodal 3D Object Detection

An Efficient Heterogeneous Edge-Cloud Learning Framework For Spectrum Data Compression

Attention-based Variable-Size Feature Compression Module for Edge Inference

Efficient On-Board Compression for Arbitrary-Shaped Cloud-Covered Remote Sensing Images via Adaptive Filling and Controllable Quantization

A Novel Image Compression Framework at Edges

An Efficient Image Compression Method Based on Cloud Data

Attention-based Feature Compression for CNN Inference Offloading in Edge Computing

A feature compression method based on similarity matching

Cloud-Edge Inference under Communication Constraints: Data Quantization and Early Exit.

Efficient Privacy Preserving Edge Computing Framework for Image Classification

Feature Compression for Rate Constrained Object Detection on the Edge

Edge-oriented Compression Coding on Image Sequence

Adaptive Compression Offloading and Resource Allocation for Edge Vision Computing

DeepCompress: Efficient Point Cloud Geometry Compression

Intermediate Deep Feature Compression: the Next Battlefield of Intelligent Sensing

EdgeCompress: Coupling Multidimensional Model Compression and Dynamic Inference for EdgeAI

CPE: an Energy-Efficient Edge-Device Training with Multi-dimensional Compression Mechanism

Adaptively Compressing IoT Data on the Resource-constrained Edge.