from IPython.display import Image as imdisplay imfile = 'sanpe300.png' imdisplay(imfile)

from PIL import Image from pprint import pprint

im = Image.open(imfile) print(type(im))

print(im.getbands()) pprint(im.info)

import numpy as np imarray = np.array(im) imarray.shape

import xarray as xr inch_mm = 25.4 def im2DataArray(im): imarray = np.array(im) dpi = im.info['dpi'] xmax_mm, ymax_mm = im.width/dpi[0]*inch_mm,im.height/dpi[1]*inch_mm xs = np.linspace(0,xmax_mm,im.width) ys = np.linspace(0,ymax_mm,im.height) da = xr.DataArray(imarray, coords = dict(y=ys,x=xs,channel = list(im.getbands())), dims = ['y','x','channel'], name='value') return da

da = im2DataArray(im) # to lower the memory footprint of imview () in some part of the notebook, I also da_low = im2DataArray(Image.open('sanpe150.png'))

from IPython.core.display import display, HTML from io import StringIO import holoviews as hv def show(h): f = StringIO() hv.renderer('bokeh').save(h,f,fmt = 'html') display(HTML(f.getvalue()))

import hvplot.xarray imview = da_low.hvplot.rgb('x','y','value',bands = 'channel',flip_yaxis=True,data_aspect=1) show(imview)

import holoviews as hv xmin,ymin,xmax,ymax = box = (22,65,32,75) square = hv.Rectangles(data=box).opts(fill_alpha=0.1) crop = da.sel(x=slice(xmin,xmax),y=slice(ymin,ymax),channel = ['R','G','B']) cropview = crop.hvplot.rgb('x','y','value',bands = 'channel',flip_yaxis=True,data_aspect=1) show((imview*square+cropview).opts(shared_axes=False).cols(1))

cropopts = dict(colorbar=False,cmap='gray',xaxis=None,yaxis=None,data_aspect=1,flip_yaxis=True) show(crop.hvplot.image('x','y',**cropopts).layout('channel'))

hist = crop.hvplot.hist('value',by='channel',alpha=0.5) show(hist)

boxes = dict() boxes[0], boxes[1] = (22,65,32,75),(22,55,32,65) square0 = hv.Rectangles(data=boxes[0]).opts(fill_alpha=0.1) square1 = hv.Rectangles(data=boxes[1]).opts(fill_alpha=0.1) selection = (imview*square0*square1) show(selection)

def boxes2crop(boxes): crops = dict() for key,box in boxes.items(): xmin,ymin,xmax,ymax = box sel = dict(x=slice(xmin,xmax),y=slice(ymin,ymax),channel = ['R']) crops[key] = da.sel(**sel).squeeze() return crops crops = boxes2crop(boxes) layout = hv.Layout([crop.hvplot.image('x','y',**cropopts,shared_axes=False) for k,crop in crops.items()]) show(layout)

from skimage.registration import phase_cross_correlation shift,_,_ = phase_cross_correlation(crops[0].data,crops[1].data,upsample_factor=10) shift

from skimage.transform import SimilarityTransform, warp t = SimilarityTransform(translation = (11.1,0)) h0 = hv.Image(crops[0].data).opts(tools = ['hover']) im1translated = warp(crops[1].data,t,preserve_range=True) h1translated = hv.Image(im1translated).opts(tools = ['hover']) delta = crops[0].data-im1translated hdelta = hv.Image(delta).opts(tools = ['hover']) show((h0+h1translated+hdelta).cols(2))

10*300/inch_mm, 11.1*inch_mm/300

translation_unit = np.array([-11.1,118.11])/16 translation_unit

vignette = da.sel(channel='R').data ims = [] for k in range(16): translation = k*translation_unit t = SimilarityTransform(translation = translation) ims.append(warp(vignette,t,preserve_range=True)) im_median = np.median(np.array(ims),axis=0) Image.fromarray(im_median>128)

boxes = dict() boxes[0], boxes[1] = (22,65,32,75),(34.085,40.915,44.085,50.915) square0 = hv.Rectangles(data=boxes[0]).opts(fill_alpha=0.1) square1 = hv.Rectangles(data=boxes[1]).opts(fill_alpha=0.1) selection = (imview*square0*square1) show(selection)

transl = np.array([-12.085,24.085]) vnorm = np.linalg.norm(transl) angle = np.arccos(transl[1]/vnorm) print('norm:',vnorm) print('angle:',angle*180/np.pi) 14./10.*vnorm/np.cos(angle),14.5/10.*vnorm/np.cos(angle)

translation_unit2 = transl/43*300/inch_mm ims = [] for k in range(43): translation = k*translation_unit2 t = SimilarityTransform(translation = translation) ims.append(warp(vignette,t,preserve_range=True)) im_median2 = np.median(np.array(ims),axis=0) Image.fromarray(im_median2>128)

from skimage import measure crop = (im_median<128)[530:600,300:1000] labels = measure.label(crop) hlabel = hv.Image(labels).opts(cmap='nipy_spectral',data_aspect=0.1,frame_height=150) show(hlabel)

subcrop = ((labels==1)*255).astype('uint8') Image.fromarray(subcrop)

def repeat(im,translation_unit,k_repeat): ims = [] for k in range(-10,k_repeat): translation = k*translation_unit t = SimilarityTransform(translation = translation) ims.append(warp(im,t,preserve_range=True)) return np.sum(np.array(ims),axis=0) curveblack = np.concatenate((np.zeros((300,700)),subcrop)).astype('uint8') curve_sum = repeat(curveblack,translation_unit,40) Image.fromarray(curve_sum.astype('uint8'))

lineblack = np.zeros(curve_sum.shape) lineblack[131:133,:] = 255 rot = SimilarityTransform(rotation = -26.6*np.pi/180,translation=(0,100)) lineblack = warp(lineblack,rot,preserve_range = True) line_sum = repeat(lineblack,translation_unit2,40) Image.fromarray(line_sum.astype('uint8'))

Image.fromarray(~((curve_sum>128) | (line_sum>128)))