These Python-based tools are developed as a POC for nabla DNA tools: mkdna.py, mkhisto.py, getdist.py and showbmp.py.
Make nabla DNA
mkdna.py [<options>] <image path or folder>
<options>
-h: help(this message)
-x <resolution>: DNA resolution (default: 4)
-d <depth>: DNA depth bit(default and max: 8)
-N: skip normalization
-S: skip nabla sum
-o <output>: save dna as an image or text
-c <threshold>: apply sobel filter(contour) with threshold(drop ratio)
if threshold > 1, pixels over threshold - 1 will be 255 gray depth.
-C <threshold>: crop area outside contour
-P <pairs file>: only make DNA's for matched pairs
Get similarity distance between two images
getdist.py [<options>] <image path> <pis path or dir> or
<image path>: get similarity of all pis pairs
<options>
-h: help(this message)
-s <minimum similarity value>: only shows pairs with larger value
-d <DNA depth>: DNA depth. If not given and not guessed, depth 8 will be assumed
-t <type>: distance type
similarity(default): cosine similarity
c-similarity(default): center weighted cosine similarity
cosine: cosine distance
euclidean: euclidean distance
histogram: histogram similarity
similarity_histo: similarity with histogram check
-P <pairs file>: only show matched pairs
Show an image bitmap with
showbmp.py [<options>] <image path>
<options>
-h: help(this message)
-x <resolution>: DNA resolution (default: 4)
-d <depth>: DNA depth bit(default and max: 8)
-m <mode>: bitmap, averaged, rotated, nabla
-r: show as raw texts
-s <scale factor>: image scaling if it is too small
-c <threshold>: apply sobel filter(contour) with threshold(drop ratio)
if threshold > 1, pixels over threshold - 1 will be 255 gray depth.
-C <threshold>: crop area outside contour
-k <kernel size>: kernel size for sobel filter
Make histogram for an image
Usage: mkhisto.py [<options>] <image or PIS path/folder>
<options>
-h: help(this message)
-x <resolution>: DNA resolution (default: 4)
-d <depth>: DNA depth bit(default and max: 8)
-N: skip normalization
-s: do nabla sum
-o <output>: save histogram as text