An important parameter when calculating the dark channel of a haze image is the size of the neighborhood. The
larger the size of the neighborhood
,
the less accurate the edge information of the dark channel , and
the more serious the Halo effect
in the dehazed image
; the smaller the size of the neighborhood, the more inaccurate the edge information of the dark channel . The more accurate the edge information is , but it is easy to cause the color of the defogged image to be oversaturated or dull. When the dark channel is obtained, when the window size of the minimum value filter is 15*15 , it is easy to blur the information on the edge of the image, but the probability of dark channels appearing in the filter window increases to make the final defogging effect more natural; while the minimum When the window size of the value filter is 3* 3 , the edge information of the image is preserved, but it will reduce the probability of dark channels appearing in the filter window, resulting in excessive defogging of the final restored image, and the guided filter proposed by He Kaiming The feature is that it can maintain the edge of the image while smoothing the non-edge of the image.
The image after dehazing will produce halo phenomenon in the place where the depth of field changes greatly, and the reason for this problem is that the minimum value filter is used in the dark channel calculation process.
In the process of finding the dark channel
,
the processing of the minimum value filter will cause the dark primary color to expand outward,
resulting in a larger expansion of the transmittance
,
so that the image after defogging will produce a "
halo
"
phenomenon on the edge part (the defogged image is whitish
).
Experimental results:
40*40 15*15 10*10 3*3
169.05 28.69 16.78 5.22
254.06 42.80 25.58 7.89
339.13 56.79 34.62 10.77
Use the dark channel prior dehazing algorithm to process the foggy image generated by the smoke generator: