adaptive thresholding #5
Description
Thank you for this fabulous work!
I would like to further understand the adaptive thresholding step by experimenting with it, but am looking for some implementation advice to get me started. In your paper, you write:
Threshold Map Estimation. Similar to the idea of adaptive thresholding, we estimate the threshold for each pixel by analyzing its neighboring pixels. The analysis can be conducted on the original image, however, in order to deal with the image noise and blur in real applications, analyzing
a downsampled image (scalar s1) is more accurate, which also brings speed benefits. Any pixel will be set to alpha if its value is less than alpha to remove pixels that are too dark. Average values are computed on a local region (window size w) on the downsampled image. To further handle the
image noise, the downsampled average map can be further downsampled (scalar s2). The final threshold map is achieved by upsampling the downsampled average map by s1 x s2 using bilinear interpolation, see Fig. 4b.
Would you be able to provide the values for these parameters, s1, s2, alpha, w? And maybe some hints on which opencv functions you use to implement the various steps? That would be immensely helpful.
Finally, would you be able to provide the input image of Figure 4 in original resolution so that i can try to reproduce Figure 4b and 4c as a check that i am understanding things correctly?
Thank you very much!