Python package

Author: epx-jgonzalez

MANIFEST.in

@@ -0,0 +1 @@
+include smartcrop/cascades/*

bin/smartcrop

@@ -0,0 +1,4 @@
+#!/usr/bin/env python
+
+import smartcrop
+smartcrop.main()

setup.py

@@ -0,0 +1,13 @@
+from setuptools import setup
+
+setup(name='smartcrop',
+      version='0.1',
+      description='OpenCV smart crop',
+      url='https://github.com/epixelic/python-smart-crop',
+      author='Josua Gonzalez',
+      author_email='[email protected]',
+      include_package_data=True,
+      license='MIT',
+      packages=['smartcrop'],
+      scripts=['bin/smartcrop'],
+      zip_safe=False)

smartcrop/__init__.py

@@ -0,0 +1,134 @@
+from __future__ import print_function
+from __future__ import division
+
+import cv2
+import argparse
+import os
+
+cascade_path = os.path.dirname(__file__) + '/cascades/haarcascade_frontalface_default.xml'
+
+def center_from_faces(matrix):
+    face_cascade = cv2.CascadeClassifier(cascade_path)
+    faces = face_cascade.detectMultiScale(matrix, 1.3, 5)
+
+    x = 0
+    y = 0
+    weight = 0
+
+    for (x, y, w, h) in faces:
+        print('Face detected at ', x, y, w, h)
+        weight += w * h
+        x += (x + w / 2) * w * h
+        y += (y + h / 2) * w * h
+
+    if len(faces) == 0:
+        return False
+
+    return {
+        'x': x / weight,
+        'y': y / weight
+    }
+
+
+def center_from_good_features(matrix):
+    x = 0
+    y = 0
+    weight = 0
+
+    max_corners = 50
+    quality_level = 0.1
+    min_distance = 10
+
+    corners = cv2.goodFeaturesToTrack(matrix, max_corners, quality_level, min_distance)
+
+    for point in corners:
+        weight += 1
+        x += point[0][0]
+        y += point[0][1]
+
+    return {
+        'x': x / weight,
+        'y': y / weight
+    }
+
+
+def crop_vertical_pos(center, original_width, original_height, target_width, target_height):
+    new_height = target_height / target_width * original_width
+    top = max(round(center['y'] - new_height / 2), 0)
+    bottom = min(top + new_height, original_height)
+
+    return {
+        'left': 0,
+        'right': original_width,
+        'top': top,
+        'bottom': bottom
+    }
+
+
+def crop_horizontal_pos(center, original_width, original_height, target_width, target_height):
+    new_width = target_width / target_height * original_height
+    left = max(round(center['x'] - new_width / 2), 0)
+    right = min(left + new_width, original_width)
+
+    return {
+        'left': left,
+        'right': right,
+        'top': 0,
+        'bottom': original_height
+    }
+
+
+def smart_crop(image, target_width, target_height, destination):
+    # read grayscale image
+    original = cv2.imread(image)
+
+    if original is None:
+        print("Could not read source image")
+        exit(1)
+
+    height, width, depth = original.shape
+
+    target_height = int(target_height)
+    target_width = int(target_width)
+
+    matrix = cv2.imread(image, cv2.CV_LOAD_IMAGE_GRAYSCALE)
+
+    ratio = max(target_width / width, target_height / height)
+
+    width, height = original.shape[1] * ratio, original.shape[0] * ratio
+    original = cv2.resize(original, (int(width), int(height)))
+    matrix = cv2.resize(matrix, (int(width), int(height)))
+
+    center = center_from_faces(matrix)
+
+    if not center:
+        print('Using Good Feature Tracking method')
+        center = center_from_good_features(matrix)
+
+    print('Found center: ', center)
+
+    if width / height < target_width / target_height:
+        crop_pos = crop_vertical_pos(center, width, height, target_width, target_height)
+    else:
+        crop_pos = crop_horizontal_pos(center, width, height, target_width, target_height)
+
+    print('Crop: ', crop_pos)
+
+    cropped = original[crop_pos['top']: crop_pos['bottom'], crop_pos['left']: crop_pos['right']]
+    cv2.imwrite(destination, cropped)
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("-W", "--width", required=True, help="Target width")
+    ap.add_argument("-H", "--height", required=True, help="Target height")
+    ap.add_argument("-i", "--image", required=True, help="Image to crop")
+    ap.add_argument("-o", "--output", required=True, help="Output")
+
+    args = vars(ap.parse_args())
+
+    smart_crop(args["image"], args["width"], args["height"], args["output"])
+
+
+if __name__ == '__main__':
+    main()