def equalize_images(images_in, e_type): images_out = imports.deepcopy(images_in) for i in range(images_out.shape[0]): arr_image = images_out[i]*255 image_to_eq = imports.array_to_img(arr_image) if (e_type == 0): images_out[i] = (imports.img_to_array(hsv_equalization(imports.np.asarray(image_to_eq))) / 255).astype('float32') if (e_type == 1): images_out[i] = (imports.img_to_array(rgb_equalization(imports.np.asarray(image_to_eq))) / 255).astype('float32') if (e_type == 2): images_out[i] = (imports.img_to_array(yuv_equalization(imports.np.asarray(image_to_eq))) / 255).astype('float32') return images_out
def load_data_eq(): meta_data = imports.pd.read_csv('data/messidor/train/messidor_annotation.csv') Y = meta_data['Retinopathy grade'].values # Transform into binary classificaiton if settings.nb_classes == 2: Y[Y > 0] = 1 n_samples = Y.shape[0] X = imports.np.empty((n_samples, settings.img_rows, settings.img_cols, 3)) for i in range(n_samples): filename = './data/messidor/train/{}'.format(meta_data['Image name'][i]) img_cv = imports.cv2.resize(imports.cv2.imread(filename), (settings.img_rows, settings.img_cols)) x = imports.img_to_array(img_cv) / 255.0 X[i] = x.astype('float32') input_shape_l = (settings.img_rows, settings.img_cols, 3) return X, Y, input_shape_l