Single image dehazing is a critical problem in computer vision. However, most recently proposed learning-based dehazing methods achieve unsatisfactory quality with dehazed images due to inaccurate parametric estimation. The size of these models is also large to be applied with mobile devices limited resources. Last, most models are tailored to image dehazing, achieving poor migration. Thus, we propose a compact multiscale attention feature fusion network with a model size of 2 MB called MSAFF-Net to perform end-to-end single image dehazing. In the proposed model, we design a simple and powerful feature extraction module to extract complex features from hazy images. We use a channel attention module and a multiscale spatial attention module to consider the regions with haze-relevant features. To our knowledge, this study is the first to directly apply the attention mechanism rather than to embed it into certain modules for single image dehazing. We compare MSAFF-Net with other approaches on the NTIRE18, RESIDE, and Middlebury Stereo datasets. We show that MSAFF-Net achieves comparable or better performance than other models. We also extend MSAFF-Net to single image deraining, and various experiments demonstrate its effectiveness. Results suggest that MSAFF-Net can directly restore clear images using channels with the most useful haze- or rain-relevant features and spatial locations.

Abstract A new two‐dimensional chaotic system in the form of a cascade structure is designed, which is derived from the Chebyshev system and the infinite collapse system. Performance analysis including trajectory, Lyapunov exponent and approximate entropy indicate that it has a larger chaotic range, better ergodicity and more complex chaotic behaviour than those of advanced two‐dimensional chaotic system recently proposed. Moreover, to protect the security of the crowd image data, the newly designed two‐dimensional chaotic system is utilized to propose a visually meaningful image cryptosystem combined with singular value decomposition and Bernstein polynomial. First, the plain image is compressed by singular value decomposition, and then encrypted to the noise‐like cipher image by scrambling and diffusion algorithm. Later, the steganographic image is obtained by randomly embedding the cipher image into a carrier image in spatial domain through the Bernstein polynomial‐based embedding method, thereby realizing the double security of image information and image appearance. Besides, the visual quality of the steganographic image can be improved by the adjustment factor according to different carrier images during the embedding process. Ultimately, security analyses indicate that it has higher encryption efficiency (2 Mbps) and the visual quality of steganography image can reach 39 dB.

This paper presents an optimization scheme for encoding the list of insignificant sets (LIS) and the list of insignificant pixels (LIP) to reduce the redundancies existing in the conventional set partitioning in hierarchical trees (SPIHT) based image compression algorithm. This scheme is based upon the investigation of distribution characteristics of wavelet coefficients. A judgment method with a predetermined threshold value is used for the sorting pass of LIS to identify significant coefficients ahead of encoding the LIS. Assuming that all the coefficients in LIS are less than the threshold value, and then they will not be scanned and encoded by the SPIHT. Moreover, a flag representing the number of significant coefficients encoded was introduced for encoding LIP in order to stop scanning LIP once all the significant coefficients have been encoded. Experimental results with various benchmark images show that the modified SPIHT (MSPIHT) achieves better visual quality and higher Peak Signal to Noise Ratio (PSNR) gains. Furthermore, the number of output 0 bits is significantly decreased at various bit rates compared with the original SPIHT.

This paper presents an efficient disparity estimation-based scheme for stereo image coding. This scheme makes use of a pseudo-disparity image instead of a true (estimated) disparity map to reconstruct the target image. Moreover, a block-partition approach is utilized to improve the reconstructed image quality of object edges. The major feature of this scheme is that not only its computational load is decreased drastically for calculating the disparity vectors compared to true disparity estimation (DE) and disparity compensation (DC)-based algorithm, but also, its reconstructed image quality is better than true DE/DC-based algorithm. Furthermore, in the proposed scheme, a reference image is compressed by lossy JPEG2000, while a pseudo disparity image is encoded by a lossless way for good reconstruction of the target image. The simulation results show that the PSNR is better than true DE/DC-based algorithm at the identical bit rate.