Integration of enhanced dictionary learning and magnitude computation techniques for removing rain streaks in digital image enhancement
Rain streaks detection and removal are very important topics in the field of image processing and computer vision. The present of rain in images and videos causing the pixels in the image corrupted. The main problem in this research is to detect and remove rain streaks. The aim of this research is t...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2017
|
| Subjects: | |
| Online Access: | https://eprints.ums.edu.my/id/eprint/38979/1/24%20PAGES.pdf https://eprints.ums.edu.my/id/eprint/38979/2/FULLTEXT.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Rain streaks detection and removal are very important topics in the field of image processing and computer vision. The present of rain in images and videos causing the pixels in the image corrupted. The main problem in this research is to detect and remove rain streaks. The aim of this research is to develop an efficient technique that able to detect and remove rain streaks from a single image. The proposed technique consists of two main parts which are the rain streaks detection and rain streaks removal. Both rain streaks detection and rain streaks removal are combined and known as HyDRa. At first, the contrast of the image will be enhanced followed by bilateral filtering technique to divide the input image into two parts, low frequency, and high-frequency part. The classification of rain component and non-rain component is done when the high-frequency part of the image undergoes the dictionary learning approach. For achieving smooth detection of the rain streaks process, the magnitude of each pixel in the rain component will be computed. As for the removal stages, the non-rain component is subtracted from the image and will be combined with the low-frequency part from the filtering stage. The PSNR test and SSiM test of HyDRa for image 1 are 31.09 dB and 0.9194 respectively. Based on the performance test, the PSNR values for test images are significantly better as compared to the classical technique such bilateral filtering approach and self-learning dictionary approach. |
|---|