Data augmentation

python/x0512.py

@@ -0,0 +1,131 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+
+### DATA 622 Week 10 Homework
+### Youqing Xiang
+### user.id: x0512
+
+"""
+# Summary
+
+# There are three main parts in this file:
+#     Part I: imageProcess function
+#     Part II: 10 methods used in image process
+#     Part III: the main function as an example of how to run the code
+
+# For imageProcess function, it starts from the legend file (image infor file),
+#     and then search for the image in image file and do image transformation.
+#     So, this function would ignore any image in image file if it is not or
+#     not correctly recorded in legend file.
+"""
+
+import os
+import pandas as pd
+from keras.preprocessing.image import ImageDataGenerator, img_to_array, load_img
+
+def imageProcess(datagen,image_dir,legend_dir):
+    """A function for image transformation.
+
+    input:
+        datagen: the method for image process
+        image_dir: the image file direction
+        legend_dir: the image infor (legend) file direction
+
+    output:
+        a dataframe: including the original image information for image process
+        transformed images: saved in the new_images folder
+    """
+    # read legend file
+    legend = pd.read_csv(legend_dir)
+    image_list = legend['image']
+    primary_list = legend['Primary']
+    secondary_list = legend['Secondary']
+    n = len(image_list)
+    n = 5
+
+    # create new_images folder for transformed images
+    if not os.path.exists('new_images'):
+        os.makedirs('new_images')
+
+    # image transformation and saving original information for transformed images
+    columns = ['user.id', 'image', 'Primary','Secondary']
+    legend_result = pd.DataFrame(columns=columns)
+
+    for i in range(0,n):
+        try:
+            img = load_img(os.path.join(image_dir, image_list[i]))
+        except:
+            print 'image reading error'
+            continue
+
+        df = pd.DataFrame([[legend['user.id'][i],legend['image'][i],
+                            legend['Primary'][i],legend['Secondary'][i]]],
+                          columns=columns)
+        legend_result = legend_result.append(df)
+
+        img = img_to_array(img)
+        img = img.reshape((1,) + img.shape)
+
+        name = legend['image'][i].split('.')[0]
+
+
+        for image in datagen.flow(img,batch_size=1,save_to_dir='new_images',
+                                  save_prefix=name):
+            break
+
+    return legend_result
+
+
+# 10 methods could be used for image process
+datagen1 = ImageDataGenerator(shear_range = 0.1)
+
+datagen2 = ImageDataGenerator(rotation_range = 5)
+
+datagen3 = ImageDataGenerator(width_shift_range = 0.1)
+
+datagen4 = ImageDataGenerator(height_shift_range = 0.1)
+
+datagen5 = ImageDataGenerator(shear_range = 0.1,
+                              rotation_range = 5)
+
+datagen6 = ImageDataGenerator(width_shift_range = 0.1,
+                              rotation_range = 5)
+
+datagen7 = ImageDataGenerator(rotation_range = 5,
+                              height_shift_range = 0.1)
+
+datagen8 = ImageDataGenerator(width_shift_range = 0.1,
+                              height_shift_range = 0.1)
+
+datagen9 = ImageDataGenerator(shear_range = 0.1,
+                              height_shift_range = 0.1)
+
+datagen10 = ImageDataGenerator(shear_range = 0.1,
+                               width_shift_range = 0.1)
+
+datagens = [datagen1,datagen2,datagen3,datagen4,datagen5,
+              datagen6,datagen7,datagen8,datagen9,datagen10]
+
+
+
+if __name__ == "__main__":
+    image_dir = 'Xiang_image'
+    legend_dir = 'Xiang_data/legend.csv'
+
+    columns = ['user.id', 'image', 'Primary','Secondary']
+    legend = pd.DataFrame(columns=columns)
+
+    for datagen in datagens:
+        df = imageProcess(datagen,image_dir,legend_dir)
+        legend = pd.concat([legend,df])
+
+    newlist = pd.Series(os.listdir('new_images'))
+    legend = legend.sort_values(by='image')
+    legend.reset_index(drop=True,inplace=True)
+
+    legend['image'] = newlist
+
+    if not os.path.exists('new_data'):
+        os.makedirs('new_data')
+
+    legend.to_csv('new_data/new_legend.csv')