Huang Jijiang

Huang Jijiang

Auto-generated profile, based on publicly available information

Huang Jijiang

Xi'an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences, China

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 December 2014
Verified
Conference Paper
Closed access
JLjiawen liao
jiawen liao
HJHuang Jijiang
Huang Jijiang
ZWZefeng Wang
Zefeng Wang
WHWang Huawei
Wang Huawei
LYLei Yang
Lei Yang
LRLong Ren
Long Ren
Pointing at the defect of Space Frequency Algorithm in abstracting detail information used in Regional Feature Self-Adaptive Image Fusion Algorithm, an improved Spatial Frequency Algorithm is proposed. Firstly, the histogram specification is used to adjust pixel values of the infrared image. Secondly, the multi-scale decomposition is performed on visible image and the adjusted infrared image using Contourlet transform. Finally, low-frequency sub-band is fused using a rule based on Regional Energy rather than the simple weighted sum method. Other sub-bands are fused using the Improve Spatial Frequency rule. The experimental results show that the proposed method preserves more details of the visible image and obtains clear target information in the infrared image. A better effect is achieved compared to the reference article.

16 December 2022
Verified
Conference Paper
Open access
LGLei Guan
Lei Guan
JHJijiang Huang
Jijiang Huang
HWHao Wang
Hao Wang
WCWeining Chen
Weining Chen
RLRong Lu
Rong Lu
ZCZhiqiang Chen
Zhiqiang Chen
Image denoising has always been an indispensable part of image processing tasks, and the quality of image preprocessing algorithms is directly related to the effect of subsequent image processing. Most existing denoising algorithms explicitly define the type and distribution of noise, for example, the artificially set additive Gaussian white noise which is based on prior knowledge. But on the one hand, the actual real image noise is not equivalent to Gaussian noise. The noise distribution of different noise types such as Gaussian, Salt and pepper noise, Rayleigh, Uniform noise is very different. It is not possible to generalize in the process of image denoising. On the other hand, images in real scenarios often have problems such as unrepeatable acquisition, lack of data information, or even only single image information, which can't support large-scale deep learning and discrimination. The above two points limit the application of most algorithms in real scenarios. In this paper, a new real image noise recognition algorithm is proposed. Aiming at the problem of single noise image and lack of prior knowledge in real scenes, the noise distribution is first learned and the image enhancement dataset is generated through dual-channel simulation in traditional and deep learning to solve the problem of lack of real image data. Then, the noise is identified by the deep learning pre-trained model, and the noise type that most closely matches the noise distribution of the real image is found. Thus, the influence of prior knowledge of image noise on the denoising process is freed for the subsequent denoising task. Experimental results show that our algorithm framework is not limited by the size of real images and datasets, and does not require prior knowledge of real scenes. The algorithm has good recognition accuracy for different types of image noise, and good robustness for images of various scenes, which provides data support and accuracy guarantee for a series of tasks such as subsequent denoising, classification, recognition, and tracking of computer vision.