preprocess_all.py

from PIL import Image
import numpy as np
import os
import glob
import csv
import multiprocessing
from multiprocessing import Pool
csv.register_dialect("textdialect", delimiter='\t')
'''

Converts images to their greyscale + cropped versions.

'''


def process_image(file, fldr):
        try: im = Image.open(file, 'r')
        except: 
               writer.writerow(["Could not open:", file])
               return
        
        try: im = im.convert('L')  # makes it greyscale
        except: 
               writer.writerow(["Could not convert to grayscale:", file])
               return
               
        width, height = im.size # resize image and crop
        if width <= height:
               new_height = int(height * (float(CROP_WIDTH)/width))
               im = im.resize((CROP_WIDTH, new_height), Image.BICUBIC)
               left, right = 0, CROP_WIDTH
               top, bottom = (new_height - CROP_HEIGHT)/2, (new_height + CROP_HEIGHT)/2
               while bottom - top < CROP_HEIGHT:
                     bottom += 1
        elif height < width:
               new_width = int(width * (float(CROP_HEIGHT)/height))
               im = im.resize((new_width, CROP_HEIGHT), Image.BICUBIC)
               top, bottom = 0, CROP_HEIGHT
               left, right = (new_width - CROP_WIDTH)/2, (new_width + CROP_WIDTH)/2
               while right - left < CROP_WIDTH:
                     right += 1
        im = im.crop((left, top, right, bottom))

        data = np.asarray(im)
        #print data
        #print data.shape
        result_width, result_height = data.shape
        filename = os.path.basename(file)
        filepath = os.path.join(fldr, '%s_postproc.jpg' % filename)
        if result_width != CROP_WIDTH or result_height != CROP_HEIGHT:
               #print 'Fail to save due to sizing error: %s' % filepath
               writer.writerow(["Sizing error", filepath])
        else:
               try:
                     out_im = Image.fromarray(data, mode='L')
                     out_im.save(filepath)
               except:
                     writer.writerow(["Could not save", filepath])
                     return
               #print 'Saved %s' % filepath


# TODO: move the following to constants.py
# DATA_DIR = '../data/toy_data_raw'
# DATA_PATH = '../data/lol'
# OUT_PATH = '../data/lol_processed'
DATA_PATH = 'G:\\MechanicalCuratorReleaseData\\extractedimagedata\\'
OUT_PATH = 'D:\\ArtHistoryNet\\images_postproc_256\\'
CROP_WIDTH, CROP_HEIGHT = 256, 256 # might want to change to 256 in future for more granularity

if __name__ == "__main__":

        os.chdir(DATA_PATH)
        try: os.mkdir(OUT_PATH)
        except: pass
        
        multiprocessing.freeze_support()
        p = Pool(4)
        
        counter = 0
        logFile = open("log_256_2.txt", 'a')
        writer = csv.writer(logFile, 'textdialect')
        for imgsize in ['plates']:
        #for imgsize in ['medium']:
               try: os.mkdir(OUT_PATH + '\\' + imgsize)
               except: pass
               
               for date in glob.glob(imgsize + '\\1*'):
                     dt = int(date[-4:])
                     if dt < 1893: continue  # that is because the previous run stopped there
                     
                     print date
                     try: os.mkdir(OUT_PATH + '\\' + date)
                     except: pass
                     
                     fldr = OUT_PATH + '\\' + date
                     for file in glob.glob(date + '\\*.jpg'):
                          if file.startswith('.') or not file.endswith('.jpg'): continue
                          filename = os.path.basename(file)
                          filepath = os.path.join(fldr, '%s_postproc.jpg' % filename)
                          if glob.glob(filepath): continue  # already done
                          
                          p.apply_async(process_image, args=(file, fldr))
                          counter += 1
                          if counter % 100 == 0: print counter
        p.close()
        p.join()