extract_and_save.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Fri Jun  3 19:29:26 2022
@license: MIT

@author: Dulfiqar 'activexdiamond' H. Al-Safi
"""

############################## Dependencies ##############################
##Path utils for image loading.
import os

##Maths
import numpy

##RNG

## Data serialization and deserialization
import deepdish

##Image manipulation.
import cv2
import skimage

##Visualization
from matplotlib import pyplot

############################## Custom Modules ##############################
import preprocessing
import sfta
import debugging 

############################## Config ##############################
import config
import debug_config

def main():
    raw_images = preprocessing.load_images(config.IMAGE_RELATIVE_PATH)
    labels = preprocessing.load_labels(config.LABEL_RELATIVE_PATH)
    
    dataset_len = len(raw_images)
    image_shape = (dataset_len, config.IMAGE_W, config.IMAGE_H)

    images = numpy.empty(image_shape)
    features = []
    
    #for i in range(3):
    #    image = raw_images[i]
    for i, image in enumerate(raw_images):
        print(f"Stripping skull from image at index: {i}")
        #The third param should be True if you wish to display plots to the user during processing.
        image = preprocessing.strip_skull(image, True, i < debug_config.SKULL_STRIPS_TO_DISPLAY, i)            
        ##Convert to gray-scale.
        image = cv2.cvtColor(image, config.FINAL_GRAYING_MODE)
        ##Crop and resize.
        image = cv2.resize(image, (config.IMAGE_SIZE))
        
        feat = {}
        ## LBP
        feat["lbp"] = preprocessing.compute_lbp(image)
        
        ## HOG
        feat["hog"], hog_image = skimage.feature.hog(image, visualize=True)
        
        ## SFTA
        feat["sfta"] = sfta.compute_sfta(image, 5)
        
        ## Third Order - GLCM
        feat["glcm"] = preprocessing.compute_glcm_stats(image)
        
        ## First Order - Stats    
        feat["fo"] = preprocessing.compute_fo_all(image)        
        
        ##Debug 
        debugging.plot_features(i, image, feat)
        
        ##Final step.
        features.append(feat)
        images[i, :, :] = image

    print(f"Saving processed images to: {config.IMAGE_OUTPUT_PATH}")
    image_path = os.getcwd() + config.IMAGE_OUTPUT_PATH
    for i, image in enumerate(images):
        filename = image_path + str(i) + ".png"
        print(i, filename)
        succ = cv2.imwrite(filename, image)
        if not succ:
            print(f"Failed to save image: {i}")
    
    print(f"Saving extracted features to: {config.FEATURE_OUTPUT_PATH}")
    feature_path = os.getcwd() + config.FEATURE_OUTPUT_PATH
    deepdish.io.save(feature_path, features)
    
    
    ############################## Debug-Plot Images
    ##Plot the proportion of normal and abnormal labels.
    pyplot.bar(labels, debug_config.PLOT_RATIO_BARS_H)
    
    ##Display the images as matplotlib plots, to get an idea of what we're working with.
    ##This is the main purpose of the d_<x>_images arrays.
    debugging.image_plotter(images, labels, debug_config.PLOT_IMAGE_FIGURE_SIZE, debug_config.PLOTTED_IMAGE_COUNT,
                            "Final Result", debug_config.PLOT_IMAGE_OFFSET)
    #debugging.image_plotter(d_raw_images, labels, debug_config.PLOT_IMAGE_FIGURE_SIZE, debug_config.PLOTTED_IMAGE_COUNT,
    #                        "Raw", debug_config.PLOT_IMAGE_OFFSET)
    #debugging.image_plotter(d_brain_images, labels, debug_config.PLOT_IMAGE_FIGURE_SIZE, debug_config.PLOTTED_IMAGE_COUNT,
    #                        "Skull-Stripped", debug_config.PLOT_IMAGE_OFFSET)
    #debugging.image_plotter(d_gray_images, labels, debug_config.PLOT_IMAGE_FIGURE_SIZE, debug_config.PLOTTED_IMAGE_COUNT,
    #                        "Grayed", debug_config.PLOT_IMAGE_OFFSET)
    

    
    
if __name__ == '__main__':
    main()