normalize.py

# -*- coding: utf-8 -*-
"""
Created on Wed Nov 21 15:16:06 2018

@author: Murali
"""

import cv2
import numpy as np
from scipy import ndimage
from scipy.misc import imresize


def preprocess_signature(img, canvas_size, img_size=(170, 242), input_size=(150, 220)):
    img = img.astype(np.uint8)
    centered = normalize_image(img, canvas_size)
    inverted = 255 - centered
    resized = resize_image(inverted, img_size)
    cropped = crop_center(resized, input_size)
    return cropped


def normalize_image(img, size=(840, 1360)):
    """ Centers an image in a pre-defined canvas size, and remove
    noise using OTSU's method.

    :param img: The image to be processed
    :param size: The desired canvas size
    :return: The normalized image
    """

    max_r, max_c = size

    # 1) Crop the image before getting the center of mass

    # Apply a gaussian filter on the image to remove small components
    # Note: this is only used to define the limits to crop the image
    blur_radius = 2
    blurred_image = ndimage.gaussian_filter(img, blur_radius)

    # Binarize the image using OTSU's algorithm. This is used to find the center
    # of mass of the image, and find the threshold to remove background noise
    threshold, binarized_image = cv2.threshold(blurred_image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

    # Find the center of mass
    r, c = np.where(binarized_image == 0)
    r_center = int(r.mean() - r.min())
    c_center = int(c.mean() - c.min())

    # Crop the image with a tight box
    cropped = img[r.min(): r.max(), c.min(): c.max()]

    # 2) Center the image
    img_r, img_c = cropped.shape

    r_start = max_r // 2 - r_center
    c_start = max_c // 2 - c_center

    # Make sure the new image does not go off bounds
    # Case 1: image larger than required (height):  Crop.
    # Emit a warning since we don't want this for the signatures in the WD dataset (OK for feature wi)
    if img_r > max_r:
        print ('Warning: cropping image. The signature should be smaller than the canvas size')
        r_start = 0
        difference = img_r - max_r
        crop_start = difference // 2
        cropped = cropped[crop_start:crop_start + max_r, :]
        img_r = max_r
    else:
        extra_r = (r_start + img_r) - max_r
        # Case 2: centering exactly would require a larger image. relax the centering of the image
        if extra_r > 0:
            r_start -= extra_r
        if r_start < 0:
            r_start = 0

    # Case 2: image larger than required (width). Crop.
    if img_c > max_c:
        print ('Warning: cropping image. The signature should be smaller than the canvas size')
        c_start = 0
        difference = img_c - max_c
        crop_start = difference // 2
        cropped = cropped[:, crop_start:crop_start + max_c]
        img_c = max_c
    else:
        extra_c = (c_start + img_c) - max_c
        if extra_c > 0:
            c_start -= extra_c
        if c_start < 0:
            c_start = 0

    normalized_image = np.ones((max_r, max_c), dtype=np.uint8) * 255
    # Add the image to the blank canvas
    normalized_image[r_start:r_start + img_r, c_start:c_start + img_c] = cropped

    # Remove noise - anything higher than the threshold. Note that the image is still grayscale
    normalized_image[normalized_image > threshold] = 255

    return normalized_image


def remove_background(img):
        """ Remove noise using OTSU's method.

        :param img: The image to be processed
        :return: The normalized image
        """

        img = img.astype(np.uint8)
        # Binarize the image using OTSU's algorithm. This is used to find the center
        # of mass of the image, and find the threshold to remove background noise
        threshold, _ = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

        # Remove noise - anything higher than the threshold. Note that the image is still grayscale
        img[img > threshold] = 255

        return img


def resize_image(image, new_size, interpolation='bilinear'):
    """ Crops an image to the desired size without stretching it.

    :param image: The image to be cropped
    :param new_size: The desired size
    :param interpolation: The interpolation method ('nearest', 'bilinear', 'bicubic')
    :return: The cropped image
    """
    height, width = new_size

    # Check which dimension needs to be cropped
    # (assuming the new height-width ratio may not match the original size)
    width_ratio = float(image.shape[1]) / width
    height_ratio = float(image.shape[0]) / height
    if width_ratio > height_ratio:
        resize_height = height
        resize_width = int(round(image.shape[1] / height_ratio))
    else:
        resize_width = width
        resize_height = int(round(image.shape[0] / width_ratio))

    # Resize the image (will still be larger than new_size in one dimension)
    image = imresize(image.astype(np.float32), (resize_height, resize_width), interp=interpolation)

    # Crop to exactly the desired new_size, using the middle of the image:
    if width_ratio > height_ratio:
        start = int(round((resize_width-width)/2.0))
        return image[:, start:start+width]
    else:
        start = int(round((resize_height-height)/2.0))
        return image[start:start+height, :]


def crop_center(img, input_shape):
    img_shape = img.shape
    start_y = (img_shape[0] - input_shape[0]) // 2
    start_x = (img_shape[1] - input_shape[1]) // 2
    cropped = img[start_y: start_y + input_shape[0], start_x:start_x + input_shape[1]]
    return cropped