论文阅读笔记（五十六）：Image Super-Resolution Using Deep Convolutional Networks

Abstract—We propose a deep learning method for single image super-resolution (SR). Our method directly learns an end-to-end mapping between the low/high-resolution images. The mapping is represented as a deep convolutional neural network (CNN) that takes the low-resolution image as the input and outputs the high-resolution one. We further show that traditional sparse-coding-based SR methods can also be viewed as a deep convolutional network. But unlike traditional methods that handle each component separately, our method jointly optimizes all layers. Our deep CNN has a lightweight structure, yet demonstrates state-of-the-art restoration quality, and achieves fast speed for practical on-line usage. We explore different network structures and parameter settings to achieve tradeoffs between performance and speed. Moreover, we extend our network to cope with three color channels simultaneously, and show better overall reconstruction quality.

Index Terms—Super-resolution, deep convolutional neural networks, sparse coding

1 INTRODUCTION
Single image super-resolution (SR) [20], which aims at recovering a high-resolution image from a single low-resolution image, is a classical problem in computer vision. This problem is inherently ill-posed since a multiplicity of solutions exist for any given low-resolution pixel. In other words, it is an underdetermined inverse problem, of which solution is not unique. Such a problem is typically mitigated by constraining the solution space by strong prior information. To learn the prior, recent state-of-the-art methods mostly adopt the example-based [46] strategy. These methods either exploit internal similarities of the same image [5], [13], [16], [19], [47], or learn mapping functions from external low and high-resolution exemplar pairs [2], [4], [6], [15], [23], [25], [37], [41], [42], [47], [48], [50], [51]. The external example-based methods can be formulated for generic image super-resolution, or can be designed to suit domain specific tasks, i.e., face hallucination [30], [50], according to the training samples provided.

The sparse-coding-based method [49], [50] is one of the representative external example-based SR methods. This method involves several steps in its solution pipeline. First, overlapping patches are densely cropped from the input image and pre-processed (e.g.,subtracting mean and normalization). These patches are then encoded by a low-resolution dictionary. The sparse coefficients are passed into a high-resolution dictionary for reconstructing high-resolution patches. The overlapping reconstructed patches are aggregated (e.g., by weighted averaging) to produce the final output. This pipeline is shared by most external example-based methods, which pay particular attention to learning and optimizing the dictionaries [2], [49], [50] or building efficient mapping functions [25], [41], [42], [47]. However, the rest of the steps in the pipeline have been rarely optimized or considered in an unified optimization framework.

In this paper, we show that the aforementioned pipeline is equivalent to a deep convolutional neural network [27] (more details in Section 3.2). Motivated by this fact, we consider a convolutional neural network that directly learns an end-to-end mapping between low and high-resolution images. Our method differs fundamentally from existing external example-based approaches, in that ours does not explicitly learn the dictionaries [41], [49], [50] or manifolds [2], [4] for modeling the patch space. These are implicitly achieved via hidden layers. Furthermore, the patch extraction and aggregation are also formulated as convolutional layers, so are involved in the optimization. In our method, the entire SR pipeline is fully obtained through learning, with little pre/post-processing.

We name the proposed model Super-Resolution Convolutional Neural Network (SRCNN)1. The proposed SRCNN has several appealing properties. First, its structure is intentionally designed with simplicity in mind, and yet provides superior accuracy2 compared with state-of-the-art example-based methods. Figure 1 shows a comparison on an example. Second, with moderate numbers of filters and layers, our method achieves fast speed for practical on-line usage even on a CPU. Our method is faster than a number of example-based methods, because it is fully feed-forward and does not need to solve any optimization problem on usage. Third, experiments show that the restoration quality of the network can be further improved when (i) larger and more diverse datasets are available, and/or (ii) a larger and deeper model is used. On the contrary, larger datasets/models can present challenges for existing example-based methods. Furthermore, the proposed network can cope with three channels of color images simultaneously to achieve improved super-resolution performance.

Overall, the contributions of this study are mainly in three aspects:

1) We present a fully convolutional neural network for image super-resolution. The network directly learns an end-to-end mapping between low and high-resolution images, with little pre/post-processing beyond the optimization.

2) We establish a relationship between our deep learning-based SR method and the traditional sparse-coding-based SR methods. This relationship provides a guidance for the design of the network structure.

3) We demonstrate that deep learning is useful in the classical computer vision problem of super-resolution, and can achieve good quality and speed.

A preliminary version of this work was presented SC / 25.58 dB SRCNN / 27.95 dB earSClie/r25.[5181dB]. The pSrReCsNeNn/t2w7.9o5 drBk adds to the initial version in significant ways. Firstly, we improve the SRCNN by introducing larger filter size in the non-linear mapping layer, and explore deeper structures by adding non-linear mapping layers. Secondly, we extend the SRCNN to process three color channels (either in YCbCr or RGB color space) simultaneously. Experimentally, we demonstrate that performance can be improved in comparison to the single-channel network. Thirdly, considerable new analyses and intuitive explanations are added to the initial results. We also extend the original experiments from Set5 [2] and Set14 [51] test images to BSD200 [32] (200 test images). In addition, we compare with a number of recently published methods and confirm that our model still outperforms existing approaches using different evaluation metrics.

5 CONCLUSION
We have presented a novel deep learning approach for single image super-resolution (SR). We show that conventional sparse-coding-based SR methods can be reformulated into a deep convolutional neural network. The proposed approach, SRCNN, learns an end-to-end mapping between low and high-resolution images, with little extra pre/post-processing beyond the optimization. With a lightweight structure, the SRCNN has achieved superior performance than the state-of-the-art methods. We conjecture that additional performance can be further gained by exploring more filters and different training strategies. Besides, the proposed structure, with its advantages of simplicity and robustness, could be applied to other low-level vision problems, such as image deblurring or simultaneous SR+denoising. One could also investigate a network to cope with different upscaling factors.

Fig. 2. Given a low-resolution image Y, the first convolutional layer of the SRCNN extracts a set of feature maps. The second layer maps these feature maps nonlinearly to high-resolution patch representations. The last layer combines the predictions within a spatial neighbourhood to produce the final high-resolution image F (Y).

Fig. 3. An illustration of sparse-coding-based methods in the view of a convolutional neural network.