Digital images in real world applications typically undergo a wide variety of quality degradations before compression or re-compression. Existing learning based codecs are typically data-driven, relying on the predefined compression pipeline with pristine or high quality images as the input. However, the images in the wild may exhibit the substantially different characteristics compared to the high quality images, casting major challenges to the learning based image coding. In this paper, we propose a robust noisy image compression framework with the blind assumption on the specific noise type and level. The specifically designed encoder decomposes the representation of visual content into two types of features, including the Features that represent the Intrinsic Content (FIC) and the Features that account for Additive Degradation (FAD). As such, beyond the philosophy of faithfully reconstructing the given image with high fidelity, only FIC needs to be compactly represented and conveyed. The principled disentanglement strategy facilitates the removal of the redundancy from multiple perspectives (e.g., spatial, channel and content), ensuring the handling of a wide variety of noisy images in the wild. Extensive experimental results show that our model can achieve superior performance in terms of the ultimate quality and exhibit the strong generalizability across images degraded by a variety of means. The proposed scheme also points out a new research avenue on learning based compression for images in the wild, which is technically challenging but desirable in practice.

When transmission medium and compression degradation are intertwined, new challenges emerge. This study addresses the problem of raindrop removal from compressed images, where raindrops obscure large areas of the background and compression leads to the loss of high-frequency (HF) information. The restoration of the former requires global contextual information, while the latter necessitates guidance for high-frequency details, resulting in a conflict in utilizing these two types of information when designing existing methods. To address this issue, we propose a novel transformer architecture that leverages the advantages of attention mechanism and HF-friendly design to effectively restore the compressed raindrop images at the framework, component, and module levels. Specifically, at the framework level, we integrate relative position multi-head self-attention and convolutional layers into the proposed low-high-frequency transformer (LHFT), where the former captures global contextual information and the latter focuses on high-frequency information. Their combination effectively resolves the issue of mixed degradation. At the component level, we utilize high-frequency depth-wise convolution (HFDC) with zero-mean kernels to improve the capability to extract high-frequency features, drawing inspiration from typical high-frequency filters like Prewitt and Sobel operators. Finally, at the module level, we introduce a low-high-attention module (LHAM) to adaptively allocate the importance of low and high frequencies along channels for effective fusion. We establish the JPEG-compressed raindrop image dataset and conduct extensive experiments on different compression rates. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods without increasing computational costs.

We propose a talking face video compression framework by implicitly transforming the temporal evolution into compact feature representation. More specifically, the temporal evolution of faces, which is complex, non-linear and difficult to extrapolate, is modelled in an end-to-end inference framework based upon very compact features. This enables the high-quality rendering of the face videos, which benefits from the learning of dense motion map with compact feature representation. Therefore, the proposed framework can accommodate ultra-low bandwidth video communication and maintain the quality of the reconstructed videos. Experimental results demonstrate that compared with the state-of-the-art video coding standard Versatile Video Coding (VVC) as well as the latest generative compression scheme Face Video-to-Video Synthesis (Face_vid2vid), the proposed scheme is superior in terms of both objective and subjective quality assessment methods.

In this paper, we propose a deep video compression method for P-frame in sub-sampled color spaces regarding the YUV420, which has been widely adopted in many state-of-art hybrid video compression standards, in an effort to achieve high compression performance. We adopt motion estimation and motion compression to facilitate the inter prediction of the videos with YUV420 color format, shrinking the total data volume of motion information. Moreover, the motion compensation module on YUV420 is cooperated to enhance the quality of the compensated frame with the consideration of the resolution alignment in the sub-sampled color spaces. To explore the cross-component correlation, the residual encoder-decoder is accompanied with two head-branches and color information fusion. Additionally, a weighted loss emphasizing more on the Y component is utilized to enhance the compression efficiency. Experimental results show that the proposed method can realize 19.82% bit rate reductions on average compared to the deep video compression (DVC) method in terms of the combined PSNR and predominant gains on the Y component.

Compactly representing the visual signals is of fundamental importance in various image/video-centered applications. Although numerous approaches were developed for improving the image and video coding performance by removing the redundancies within visual signals, much less work has been dedicated to the transformation of the visual signals to another well-established modality for better representation capability. In this paper, we propose a new scheme for visual signal representation that leverages the philosophy of transferable modality. In particular, the deep learning model, which characterizes and absorbs the statistics of the input scene with online training, could be efficiently represented in the sense of rate-utility optimization to serve as the enhancement layer in the bitstream. As such, the overall performance can be further guaranteed by optimizing the new modality incorporated. The proposed framework is implemented on the state-of-the-art video coding standard (i.e., versatile video coding), and significantly better representation capability has been observed based on extensive evaluations.

Block-wise intra/inter prediction, transformation and quantization used in block-based hybrid video coding will inevitably result in blocking artifacts, especially at the low bit rate. To address this problem, this paper employs a deep convolutional neural network (CNN) to approximate the reverse function of video compression, motived by the great success of deep learning in computer vision fields recently. The proposed method establishes an end-to-end mapping, represented as the CNN, which takes the decompressed frame as input and outputs the enhanced one. Employing numerous sequences compressed by H.264 and HEVC reference software, the proposed CNN learns the connections between the lossy frame and the original one in an implicit way under different quantization parameters (QP). Figure 1 shows the architecture of our CNN and the pipeline of the network training. We build our network with convolution layers and ReLU layer and the weights and biases of all the convolution layers in our model are updated by minimizing the loss using stochastic gradient descent with the standard backpropagation. We implement the CNN as a post-loop deblocking filter and explore varying CNN parameters for different QPs. Various experimental results demonstrate that the proposed method is able to significantly improve the quality of enhanced frames in terms of both objective and subjective criterions.