Low Light Image Enhancement and Saliency Object Detection
- Publication Type:
- Thesis
- Issue Date:
- 2022
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Low light images represent a series of image types with great potential. Their research focuses on images and videos of the environment at dusk and near darkness. It can be widely used in night safety monitoring, license plate recognition, night scene shot, special target recognition at dusk, and other emergency events that occur under light scenes. After the environment is enhanced and combined with other tasks in computer vision and pattern recognition, it can bring many results, such as saliency detection and object detection under low illumination, and abnormal detection in crowded places under low-light environment. For the enhancement of low light and low light scenes, using traditional methods often results in over-exposure and halo conditions. Therefore, using deep learning network technology can fix and improve these specific shortcomings. For low light image enhancement, a series of qualitative and quantitative experimental comparisons conducted on a benchmark dataset demonstrate the superiority of our approach, which overcomes the drawbacks of white and colour distortion. At present, most of the research works on visual saliency have concentrated on the field of visible light, and there are few studies on night scenes. Due to insufficient lighting conditions in night scenes, and relatively lower contrasts and signal-to-noise ratios, the effectiveness of available visual features is greatly reduced. Moreover, without sufficient depth information, many features and clues are lost in the original images. Therefore, the detection of salient targets in night scenes is also difficult and it is a focus of current research in the field of computer vision. The performance leads to vague effects when the existing methods are directly con-ducted, so we adopt a new “enhance firstly, detection secondly” mechanism that firstly enhances the low-light images in order to improve the contrast and visibility, and then combines it with relevant saliency detection methods with depth information. Furthermore, we concern about the feature aggregation schemes for deep RGB-D saliency object detection and propose novel feature aggregation methods. Meanwhile, for the monocular vision, of which the depth information is hard to acquire, a novel RGB-D image saliency detection method is proposed to leverage depth cues for enhancing the saliency detection performance but without actually using depth data. The model not only outperforms the state-of-the-art RGB saliency models, but also achieves comparable or even better results compared with the state-of-the-art RGB-D saliency models.
Please use this identifier to cite or link to this item: