In this paper, a sparse binocular fusion convolution neural network is proposed to evaluate the quality of stereo image. In order to simulate the long-term fusion and processing of the left and right views in the brai...
In this paper, a sparse binocular fusion convolution neural network is proposed to evaluate the quality of stereo image. In order to simulate the long-term fusion and processing of the left and right views in the brain visual pathway, the network combines the two views of the stereo image four times, and the information processing is carried out through convolution together with the fusion operation. In addition, in order to overcome the computational-intensive and memory-intensive problems of convolution neural networks, a structural sparsity learning (SSL) method is used to regularize the proposed convolution neural network. The experimental results demonstrate that our proposed method performs effectively and efficiently. And the proposed method can achieve 2.0 × speedups on LIVE I database and 2.3× speedup on LIVE II database on the basis of improved performance.
Quality enhancement (QE) is an important post-processing technology for high-resolution video services at low bit rates, which can effectively improve the quality of compressed video. The application of deep learning ...
详细信息
ISBN:
(数字)9781728161365
ISBN:
(纸本)9781728161372
Quality enhancement (QE) is an important post-processing technology for high-resolution video services at low bit rates, which can effectively improve the quality of compressed video. The application of deep learning methods to the quality enhancement task has achieved great success in the past few years. However, the existing schemes are usually coding-independent, which still leaves room for further development of related technologies. Therefore, in this paper, we propose a quality enhancement-oriented video coding scheme. By analyzing the features of different video regions, a deep reinforcement learning model is used to determine the distortions of regions. Then during the video reconstruction, convolutional neural network (CNN)-based quality enhancement networks with different scales are selected to improve the video quality according to the distortion of different regions. Experimental results show that the proposed scheme outperforms the HEVC anchor in case of bits saving, bits allocation, and shows good visual quality especially at low bit rates.
Aiming at the online automatic target recognition for SAR images, a visual attention based scale analysis method is introduced for the feature extraction and the classifiers construction. By improving the features ada...
详细信息
This paper proposes a novel image steganography algorithm for color image. Recently, colorization-based image coding technique has been studied. In order to compress the color image effectively, this technique transfo...
This paper proposes a novel image steganography algorithm for color image. Recently, colorization-based image coding technique has been studied. In order to compress the color image effectively, this technique transform the chrominance image to a vector in a low-dimensional subspace via the colorization matrix. This paper utilizes the colorization-based image coding for steganography algorithm, where the secret data is embedded into the null space of the colorization matrix. Because the null space is high dimension enough, a large capacity data can be embedded. Numerical examples show that the proposed algorithm embeds large capacity secret data such as grayscale image into color image effectively.
We proposed a new post-equalization method of Laplacian of Gaussian regularizing for underwater visual light communication (UVLC). The experimental results show an 80% reduction of calculation resources comparing with...
详细信息
ISBN:
(纸本)9781943580705
We proposed a new post-equalization method of Laplacian of Gaussian regularizing for underwater visual light communication (UVLC). The experimental results show an 80% reduction of calculation resources comparing with traditional ISFA.
In this study a new method is proposed for inserting advertisement visuals into images automatically and without disturbing the image content. In this method important areas are determined using deep learning based ob...
详细信息
ISBN:
(纸本)9781538615010
In this study a new method is proposed for inserting advertisement visuals into images automatically and without disturbing the image content. In this method important areas are determined using deep learning based object, face and text detection, edge and saliency maps are obtained, and these information are used for the identification of the best location for inserting the advertisement visual. In order to select the best available advertisement visual from an advertisement pool shape and color features are utilized.
With the explosive increase of image data, the efficiency of both image compression and retrieval becomes unprecedent-edly significant. However, these two tasks are usually isolated executed, which waste great computa...
With the explosive increase of image data, the efficiency of both image compression and retrieval becomes unprecedent-edly significant. However, these two tasks are usually isolated executed, which waste great computational resources in large-scale image applications. In this work, we propose a joint framework called CodedRetrieval, which can find a general feature expression for both compression and retrieval based on neural network. Additionally, a two stage training strategy is designed to achieve better balance between the two distinct tasks. Experimental results show that our method can achieve competitive performance on both compression and retrieval comparing to classic methods, while saving great amount of computation time.
This paper presents an open-source software implementation for real-time 360-degree video stitching. To ensure a seamless stitching result, cylindrical and content-preserving warping are implemented to dynamically cor...
详细信息
This paper presents an open-source software implementation for real-time 360-degree video stitching. To ensure a seamless stitching result, cylindrical and content-preserving warping are implemented to dynamically correct image alignment and parallax, which may drift due to scene changes, moving objects, or camera movement. Depth variation, color changes, and lighting differences between adjacent frames are also smoothed out to improve visual quality of the panoramic video. The system is benchmarked with six 1080p videos, which are stitched into 4096×732 pixel output format. The proposed algorithm attains an output rate of 18 frames per second on GeForce GTX 1070 GPU and realtime speed can be met with a high-end GPU.
image retargeting is the technique to display images via devices with various aspect ratios and sizes. Traditional content-aware retargeting methods rely on low-level features to predict pixel-wise importance and can ...
image retargeting is the technique to display images via devices with various aspect ratios and sizes. Traditional content-aware retargeting methods rely on low-level features to predict pixel-wise importance and can hardly preserve both the structure lines and salient regions of the source image. To address this problem, we propose a novel adaptive image warping approach which integrates with deep convolutional neural network. In the proposed method, a visual importance map and a foreground mask map are generated by a pre-trained network. The two maps and other constraints guide the warping process to yield retargeted results with less distortions. Extensive experiments in terms of visual quality and a user study are carried out on the widely used RetargetMe dataset. Experimental results show that our method outperforms current state-of-art image retargeting methods.
At present, most flower images could only be recognized but not detected. They can only be used in the scenes with a single target instead of the scenes with two or more targets. Some application scenarios require the...
详细信息
暂无评论