Fast Image Dehazing Algorithm using Morphological Reconstruction

Presentation on theme: "Fast Image Dehazing Algorithm using Morphological Reconstruction"— Presentation transcript:

1 Fast Image Dehazing Algorithm using Morphological Reconstruction
Source: IEEE Transactions on Image Processing, Vol. 28, No. 5, May 2019. Author: Sebastian Salazar-Colores, Eduardo Cabal-Yepez, Juan M. Ramos-Arreguin, Guillermo Botella, Luis M. Ledesma-Carrillo and Sergio Ledesma Speaker: Hsin-Yu Lee Date: 2019/04/04

2 Outline Introduction Related works Proposed scheme Experiment results
Conclusions

3 License plate recognition
Introduction Object detection Outdoor monitoring License plate recognition Dehazing Speed

4 Related works (1/4) Optical scattering model Atmospheric light
Original image (Haze image) Result (Haze-Free Image) Transmission map

5 Related works (2/4) Dark channel prior (DCP) atmospheric light
Original image(64*64) Min channel (128,100,50) (200,100,20) (100,107,30) (40,50,50) (30,150,200) (212,111,90) (120,220,150) (114,150,30) (110,121,20) (230,120,20) (250,210,30) (90,211,30) (21,100,150) (35,120,10) (130,26,21) (203,100,10) (130,120,211) (111,15,50) (120,35,40) (37,105,50) (155,121,40) (177,110,150) (120,100,30) (140,130,40) (20,130,15) (R , G , B) 50 20 30 40 90 120 21 10 15 35 37 110 Min filter 50 20 30 90 120 21 50 20 30 90 21 K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” IEEE Transactions on Pattern Analysis and Machine Intelligence., vol. 33, no. 12, pp. 2341–2353, Dec

6 Related works (3/4) - Morphological reconstruction
Geodesic dilation Marker Geodesic dilation 形態學 主要用在黑白影像 將雜訊去除 強化影像特徵 Mask

7 Related works (4/4) - Morphological reconstruction
Geodesic erosion Marker Geodesic erosion Mask

8 Related works (5/5) - Morphological reconstruction
erosion dilation Opening Original image Opening Closing Original image

9 Proposed scheme (1/2) - Flowchart
Atmospheric light Dark channel A Morphological process

10 Proposed scheme (2/2) - Morphological process
Initial transmission map closing opening Closing->先膨脹 再侵蝕 ->先將區域旁邊的小雜訊合併再一起，但是遠離區域的雜訊是孤立的 因此再由侵蝕將其濾掉 依樣可以使輪廓平滑 不同的是 他可以把缺口連起來 Opening->先侵蝕 再膨脹 ->先將小區塊的雜訊濾掉 再將輪廓便平滑

11 Experiment results (1/4)
Qualitative comparison

12 Experiment results (2/4)
Quantitative comparison

13 Experiment results (3/4)
Quantitative comparison

14 Experiment results (3/4)
The performance analysis

15 Conclusions High speed High performance