Enhanced detail & dehaze technique (DDE) for static haze image improvement

• المؤلف الرئيسي: Danny, Yao Ngo Lung
• التنسيق: أطروحة
• اللغة: الإنجليزية; الإنجليزية
• منشور في: 2018
• الموضوعات: TA1501-1820 Applied optics. Photonics
• الوصول للمادة أونلاين: https://eprints.ums.edu.my/id/eprint/43335/1/24%20PAGES.pdf; https://eprints.ums.edu.my/id/eprint/43335/2/FULLTEXT.pdf

This research is related to the removal of the haze effects from a haze degraded image and enhances the image details. The Enhanced Detail & Dehaze technique (DDE) is proposed in order to satisfy the aim of developing an enhanced method by integrating modified channel prior scheme and enhanced Son method. The development of modified channel prior is inspired by the dark channel prior and bright channel prior where the modified channel prior is proposed as a prior that there are many dark pixels and bright pixels exist in the haze-free outdoor image. The haze removal using modified channel prior scheme is able to remove the haze effects from the haze degraded image. On the other hand, the enhanced Son method proposed by enhancing the Son method in order to overcome the limitations of the Son method, such as the brightness reversal, increasing noise and the problem of sky regions segmented into heterogeneous regions. The enhanced Son method is able to enhance the image detail and provides a level of detail control. Both modified channel prior scheme and enhanced Son method are integrated to form the proposed DDE method. The DDE method computes the modified channels based on the modified channel prior and estimates the global atmospheric light by selecting the brightest pixel among the modified channels. The transmission will be estimated with the global atmospheric light and modified channels. After that, a dehazed image will be obtained by solving the atmospheric scattering model. An image can be decomposed into base layer and detail layer. The DDE method smoothes the input image with guided image filter where the box filter contained inside had replaced by Gaussian smoothing. Later, the detail layer is obtained by subtracting the smooth image from the input image. A non-sky detail layer is proposed as the combination of the detail layer and the transmission. After that, the non-sky detail layer will be recombined with the dehazed image based on the tone transform model. The recovered image is then post processed based on the histogram equalization for the contrast enhancement. Based on the MSE and PSNR test, the DDE method obtained a better result in average as the MSE value is lower and the PSNR is higher compared to the Gibson method and the dark channel prior scheme. For instance, the sample image 1 shows the 17.1142 dB for the dark channel prior scheme, 15.0476 dB for Gibson method and 17.4007 dB for the DDE method in the PSNR test. Based on the MSSIM test, the DDE method indicated the structure of the recovered image is different with the original one compared with the other previous works. Overall, the proposed DDE method is able to remove the haze effects and enhance the image details. The DDE method also overcome the limitations of the Son method, dark look, and block or halos effects.