SF-CNN: A Fast Compression Artifacts Removal via Spatial-To-Frequency Convolutional Neural Networks

Kim, Taeoh; Lee, Hyeongmin; Son, Hanbin; Lee, Sangyoun

doi:10.1109/icip.2019.8803503

Cited by 17 publications

(12 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This paper extends our previous conference paper [26] with the following additional contributions: 1) we generalize the task-specific spatial-to-frequency networks into the family of SPNs, 2) we refactor the network details, training dataset, and training algorithms to increase the performance, 3) we add additional analysis of memory-computation tradeoffs, 4) we propose a novel SBA module, and 5) we generalize the SPNs for diverse image restoration tasks, such as color image denoising and image enhancement in various datasets.…”

Section: Introductionsupporting

confidence: 85%

“…This is because only two 1×1×B 2 convolutional layers are added in the frequency mode. Different from our previous version [26] (Config. 1), in SPN, the basis for DCT and IDCT are finetuned during training.…”

Section: ) Spatial-to-frequency Networkmentioning

confidence: 67%

“…As described in our previous work [26], image compression artifacts occur due to block-wise quantization errorr. For this task, we use another variant called the spatial-tofrequency network which has slightly better performance than the SPNs.…”

Section: Spatial-to-frequency Networkmentioning

confidence: 88%

See 2 more Smart Citations

Block-Attentive Subpixel Prediction Networks for Computationally Efficient Image Restoration

et al. 2021

Self Cite

View full text Add to dashboard Cite

Image restoration based on the Deep Convolutional Neural Network (CNN) based image restoration has demonstrated promising results in many sub-tasks, such as image super-resolution, compression artifacts removal, image denoising, and image enhancement. Compared to many CNN-based high-level vision tasks that predict sparse probabilities of each class, the CNN for image restoration requires dense pixel-level predictions with precise intensity-level values. Therefore, a minimum number of spatial pooling (or down-sampling) operations is required to maintain the image details. Therefore, designing a fast or lightweight model for image restoration is a difficult problem and is even more critical when the spatial resolution of an image becomes larger. In this paper, we propose a family of networks called Subpixel Prediction Networks (SPNs) that predict reshaped and spatially down-sampled block-wise tensors instead of raw images with full resolution. Under this scheme, spatial downsampling decreases the restoration performance less while making the network faster. We propose a novel Subpixel Block Attention (SBA) module that re-calibrates blockwise features to diminish blockwise discontinuity to increase the performance further. The experimental results reveal that these networks demonstrate good trade-offs between speed (number of computations) and restoration performance in the three image restoration tasks: image compression artifacts removal, color image denoising, and image enhancement.INDEX TERMS Image restoration, image compression artifacts removal, image denoising, image enhancement, fast deep neural networks.

show abstract

Section: Introductionsupporting

confidence: 85%

Section: ) Spatial-to-frequency Networkmentioning

confidence: 67%

See 1 more Smart Citation

Block-Attentive Subpixel Prediction Networks for Computationally Efficient Image Restoration

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

“…The dual-domain approach was further incorporated in the DMCNN model, which was based on an auto-encoder style architecture and used multi-scale loss [17]. Furthermore, Galteri et al used a generative adversarial network for artifact removal [30] while the SFCNN architecture [10] provided improvement in performance at reduced computational costs.…”

Section: Related Work a Compression Artifact Removalmentioning

confidence: 99%

Evolutionary Algorithm Based Residual Block Search for Compression Artifact Removal

Shah¹

2020

Preprint

View full text Add to dashboard Cite

<div>Lossy image compression is ubiquitously used for</div><div>storage and transmission at lower rates. Among the existing</div><div>lossy image compression methods, the JPEG standard is the most widely used technique in the multimedia world. Over the years, numerous methods have been proposed to suppress the compression artifacts introduced in JPEG-compressed images. However, all current learning-based methods include deep convolutional neural networks (CNNs) that are manually-designed by researchers. The network design process requires extensive computational resources and expertise. Focusing on this issue, we investigate evolutionary search for finding the optimal residual block based architecture for artifact removal. We first define a residual network structure and its corresponding genotype representation used in the search. Then, we provide details</div><div>of the evolutionary algorithm and the multi-objective function</div><div>used to find the optimal residual block architecture. Finally, we present experimental results to indicate the effectiveness of our approach and compare performance with existing artifact removal networks. The proposed approach is scalable and portable to numerous low-level vision tasks.</div>

show abstract

“…Some prior techniques augment standard codecs with neural networks. Neural preprocessing addresses denoising [4,5,6], and neural postprocessing is commonly used to reduce blocking and other coding artifacts [7,8,9]. In particular, Kim et al [10] improve the output of the VVC/H266 intraframe codec using residual dense networks (RDNs) and generative adversarial networks (GANs) to win the CVPR 2020 Learned Image Compression Challenge.…”

Section: Introductionmentioning

confidence: 99%

Sandwiched Image Compression: Wrapping Neural Networks Around A Standard Codec

Guleryuz

Chou

Hoppe

et al. 2021

2021 IEEE International Conference on Image Processing (ICIP)

View full text Add to dashboard Cite

We sandwich a standard image codec between two neural networks: a preprocessor that outputs neural codes, and a postprocessor that reconstructs the image. The neural codes are compressed as ordinary images by the standard codec. Using differentiable proxies for both rate and distortion, we develop a rate-distortion optimization framework that trains the networks to generate neural codes that are efficiently compressible as images. This architecture not only improves ratedistortion performance for ordinary RGB images, but also enables efficient compression of alternative image types (such as normal maps of computer graphics) using standard image codecs. Results demonstrate the effectiveness and flexibility of neural processing in mapping a variety of input data modalities to the rigid structure of standard codecs. A surprising result is that the rate-distortion-optimized neural processing seamlessly learns to transport color images using a single-channel (grayscale) codec.

show abstract

SF-CNN: A Fast Compression Artifacts Removal via Spatial-To-Frequency Convolutional Neural Networks

Cited by 17 publications

References 10 publications

Block-Attentive Subpixel Prediction Networks for Computationally Efficient Image Restoration

Block-Attentive Subpixel Prediction Networks for Computationally Efficient Image Restoration

Evolutionary Algorithm Based Residual Block Search for Compression Artifact Removal

Sandwiched Image Compression: Wrapping Neural Networks Around A Standard Codec

Contact Info

Product

Resources

About