common-augmentation.py

# -*- coding: utf-8 -*-
"""
Created on Tue Dec 19 22:17:51 2017

@author: neusoft
"""

# ############################
# #使用skimage 数据增强处理的验证 试验

from skimage import io,transform
from skimage import color,data_dir,data
from skimage import exposure,img_as_float

import numpy as np
import os

CUR_PATH = './x-image2'

SAVE_PATH = './x-gen'

GEN_PATH = './x-gen'

# ############################
# # resize
# RESIZE_HEIGHT = 256
# RESIZE_WEIGHT = 256

# # x-image resize
RESIZE_HEIGHT = 64
RESIZE_WEIGHT = 64

def img_resize(imgfile):
    img = io.imread(imgfile)
    resized_image = transform.resize(img,(RESIZE_HEIGHT,RESIZE_WEIGHT))
    return resized_image

def resize_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_resize)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\rs_' + np.str(i)+'.jpg',coll[i])

# ############################
# # gray 转换成灰度图

def img_gray_resize(imgfile):
    rgb = io.imread(imgfile)
    # 将rgb图片转换成灰度图
    gray=color.rgb2gray(rgb)
    resized_image = transform.resize(gray,(RESIZE_HEIGHT,RESIZE_WEIGHT))
    # io.imshow(resized_image)
    return resized_image

def img_gray(imgfile):
    rgb = io.imread(imgfile)
    # 将rgb图片转换成灰度图
    gray=color.rgb2gray(rgb)
    return gray

def grayandresize_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg:'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_gray_resize)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\gy_rs_' + np.str(i)+'.jpg',coll[i])

def gray_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg:'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_gray)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\gy_' + np.str(i)+'.jpg',coll[i])

# ############################
# # rescale 按比例缩放

# #高宽同比例缩放1/4
RESCALE_HEIGHT_41 = 0.25
RESCALE_WEIGHT_41 = 0.25

# #高宽同比例缩放一半
RESCALE_HEIGHT_42 = 0.5
RESCALE_WEIGHT_42 = 0.5

# #高宽同比例缩放0.8倍
RESCALE_HEIGHT_43 = 0.8
RESCALE_WEIGHT_43 = 0.8

# #高宽同比例缩放2倍
RESCALE_HEIGHT_48 = 2.0
RESCALE_WEIGHT_48 = 2.0

def img_rescale41(imgfile):
    img = io.imread(imgfile)
    resized_image = transform.rescale(img,(RESCALE_HEIGHT_41,RESCALE_HEIGHT_41))
    return resized_image

def img_rescale42(imgfile):
    img = io.imread(imgfile)
    resized_image = transform.rescale(img,(RESCALE_HEIGHT_42,RESCALE_WEIGHT_42))
    return resized_image

def img_rescale43(imgfile):
    img = io.imread(imgfile)
    resized_image = transform.rescale(img,(RESCALE_HEIGHT_43,RESCALE_WEIGHT_43))
    return resized_image

def img_rescale48(imgfile):
    img = io.imread(imgfile)
    resized_image = transform.rescale(img,(RESCALE_HEIGHT_48,RESCALE_WEIGHT_48))
    return resized_image

# #rescaleflag = 1 缩放1/4,2 缩放一半,3 缩放0.8倍,4 缩放2倍
def rescale_from_dir(imagepath,savepath,rescaleflag=2):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    if(rescaleflag == 4):
        coll = io.ImageCollection(str,load_func=img_rescale48)
    if(rescaleflag == 3):
        coll = io.ImageCollection(str,load_func=img_rescale43)
    
    if(rescaleflag == 1):
        coll = io.ImageCollection(str,load_func=img_rescale41)
    else:
        coll = io.ImageCollection(str,load_func=img_rescale42)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\rsl_' + np.str(rescaleflag)+ '_' + np.str(i)+'.jpg',coll[i])


# ############################
# # noise 随机加噪
NOISE_NUMBER = 1000

def img_noise(imgfile):
    img = io.imread(imgfile)
    height,weight,channel = img.shape
    
    #随机生成5000个椒盐噪声
    for i in range(NOISE_NUMBER):
        x = np.random.randint(0,height)
        y = np.random.randint(0,weight)
        img[x ,y ,:] = 255
    
    return img

def noise_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_noise)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\ns_' + np.str(i)+'.jpg',coll[i])

# ############################
# # flip 水平 垂直 反转图像
def img_flip_V(imgfile):
    img = io.imread(imgfile)
    vertical_flip_img = img[::-1,:,:]
    return vertical_flip_img

def img_flip_H(imgfile):
    img = io.imread(imgfile)
    horizontal_flip_img = img[:,::-1,:]
    return horizontal_flip_img

# # 垂直 反转图像
def flipV_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_flip_V)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\fv_' + np.str(i)+'.jpg',coll[i])

# # 水平 反转图像
def flipH_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_flip_H)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\fh_' + np.str(i)+'.jpg',coll[i])

# ############################
# # rotate 旋转图像 逆时针
from keras.preprocessing import image as ksimage

def ks_rotate(x, theta, row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest', cval=0.):
    rotation_matrix = np.array([[np.cos(theta), -np.sin(theta), 0],[np.sin(theta), np.cos(theta), 0],[0, 0, 1]])
    h, w = x.shape[row_axis], x.shape[col_axis]
    transform_matrix = ksimage.transform_matrix_offset_center(rotation_matrix, h, w)
    x = ksimage.apply_transform(x, transform_matrix, channel_axis, fill_mode, cval)
    return x

def img_rotate0030(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(0, 30)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate3060(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(30, 60)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate6090(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(60, 90)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate90140(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(90, 140)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate140180(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(140, 180)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate4000(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(-40, 0)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate9040(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(-90, -40)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate14090(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(-140, -90)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

def img_rotate180140(imgfile):
    img = io.imread(imgfile)    
    rotate_limit=(-180, -140)
    theta = np.pi / 180 * np.random.uniform(rotate_limit[0], rotate_limit[1]) #逆时针旋转角度
    #rotate_limit= 30 #自定义旋转角度
    #theta = np.pi /180 *rotate_limit #将其转换为PI
    img_rot = ks_rotate(img, theta)
    return img_rot

# #rotateflag = 1,2,3,4,5,6,7,8,9
# # 1[0,30],2[30,60],3[60,90],4[90,140],5[140,180],6[-180,-140],7[-140,-90],8[-90,-40],9[-40,0]
def rotate_from_dir(imagepath,savepath,rotateflag=2):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    if(rotateflag > 4):
            if(rotateflag == 5):
                coll = io.ImageCollection(str,load_func=img_rotate140180)
            if(rotateflag == 6):
                coll = io.ImageCollection(str,load_func=img_rotate180140)
            if(rotateflag == 7):
                coll = io.ImageCollection(str,load_func=img_rotate14090)
            if(rotateflag == 8):
                coll = io.ImageCollection(str,load_func=img_rotate9040)
            else:
                coll = io.ImageCollection(str,load_func=img_rotate4000)
    else:
        if(rotateflag == 4):
            coll = io.ImageCollection(str,load_func=img_rotate90140)
        if(rotateflag == 3):
            coll = io.ImageCollection(str,load_func=img_rotate6090)
        if(rotateflag == 1):
            coll = io.ImageCollection(str,load_func=img_rotate0030)
        else:
            coll = io.ImageCollection(str,load_func=img_rotate3060)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\rs_' + np.str(rotateflag) + '_' + np.str(i)+'.jpg',coll[i])

# ############################
# # shift 平移 图像

def ks_shift(x, wshift, hshift, row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest', cval=0.):
    h, w = x.shape[row_axis], x.shape[col_axis] #读取图片的高和宽
    tx = hshift * h #高偏移大小，若不偏移可设为0，若向上偏移设为正数
    ty = wshift * w #宽偏移大小，若不偏移可设为0，若向左偏移设为正数
    translation_matrix = np.array([[1, 0, tx],[0, 1, ty],[0, 0, 1]])
    transform_matrix = translation_matrix  
    x = ksimage.apply_transform(x, transform_matrix, channel_axis, fill_mode, cval)
    return x

def img_shiftleftup(imgfile):
    img = io.imread(imgfile)
    w_limit=(0.2, 0.4)
    h_limit=(0.2, 0.4)
    wshift = np.random.uniform(w_limit[0], w_limit[1])
    hshift = np.random.uniform(h_limit[0], h_limit[1])
    #wshift = 0.1 #自定义平移尺寸
    #hshift = 0.1 #自定义平移尺寸
    img_shift = ks_shift(img, wshift, hshift)
    return img_shift

def img_shiftrightup(imgfile):
    img = io.imread(imgfile)
    w_limit=(0.4, 0.8)
    h_limit=(0.4, 0.8)
    wshift = np.random.uniform(w_limit[0], w_limit[1])
    hshift = np.random.uniform(h_limit[0], h_limit[1])
    img_shift = ks_shift(img, wshift, hshift)
    return img_shift

def img_shiftuleftdown(imgfile):
    img = io.imread(imgfile)
    w_limit=(-0.4, -0.2)
    h_limit=(-0.4, -0.2)
    wshift = np.random.uniform(w_limit[0], w_limit[1])
    hshift = np.random.uniform(h_limit[0], h_limit[1])
    img_shift = ks_shift(img, wshift, hshift)
    return img_shift

def img_shiftrightdown(imgfile):
    img = io.imread(imgfile)
    w_limit=(0.4, 0.8)
    h_limit=(-0.4, -0.8)
    wshift = np.random.uniform(w_limit[0], w_limit[1])
    hshift = np.random.uniform(h_limit[0], h_limit[1])
    img_shift = ks_shift(img, wshift, hshift)
    return img_shift

# #shiftflag = 1 leftup ,2 leftdown ,3 rightup ,4 rightdown
def shift_from_dir(imagepath,savepath,shiftflag=2):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    if(shiftflag == 4):
        coll = io.ImageCollection(str,load_func=img_shiftrightdown)
    if(shiftflag == 3):
        coll = io.ImageCollection(str,load_func=img_shiftuleftdown)
    
    if(shiftflag == 1):
        coll = io.ImageCollection(str,load_func=img_shiftleftup)
    else:
        coll = io.ImageCollection(str,load_func=img_shiftrightup)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\sf_' + np.str(shiftflag) + '_' + np.str(i)+'.jpg',coll[i])

# ############################
# # zoom 缩放 图像 保持中心不变
def ks_zoom(x, zx, zy, row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest', cval=0.):
    zoom_matrix = np.array([[zx, 0, 0],[0, zy, 0],[0, 0, 1]])
    h, w = x.shape[row_axis], x.shape[col_axis]
    transform_matrix = ksimage.transform_matrix_offset_center(zoom_matrix, h, w) #保持中心坐标不改变
    x = ksimage.apply_transform(x, transform_matrix, channel_axis, fill_mode, cval)
    return x

def img_zoom41(imgfile):
    img = io.imread(imgfile)
    zoom_range=(0.1, 0.3)
    zx, zy = np.random.uniform(zoom_range[0], zoom_range[1], 2)
    
    #zx = 0.5
    #zy = 0.5 #自定义zoom尺寸
    img_zoom = ks_zoom(img, zx, zy)
    return img_zoom

def img_zoom42(imgfile):
    img = io.imread(imgfile)
    zoom_range=(0.4, 0.6)
    zx, zy = np.random.uniform(zoom_range[0], zoom_range[1], 2)
    img_zoom = ks_zoom(img, zx, zy)
    return img_zoom

def img_zoom43(imgfile):
    img = io.imread(imgfile)
    zoom_range=(0.6, 0.8)
    zx, zy = np.random.uniform(zoom_range[0], zoom_range[1], 2)
    img_zoom = ks_zoom(img, zx, zy)
    return img_zoom

def img_zoom48(imgfile):
    img = io.imread(imgfile)
    zoom_range=(1.5, 2)
    zx, zy = np.random.uniform(zoom_range[0], zoom_range[1], 2)
    img_zoom = ks_zoom(img, zx, zy)
    return img_zoom

# #zoomflag = 1 缩0.2 ,2 缩0.5 ,3 缩0.8 ,4 缩2倍
def zoom_from_dir(imagepath,savepath,zoomflag=2):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    if(zoomflag == 4):
        coll = io.ImageCollection(str,load_func=img_zoom48)
    if(zoomflag == 3):
        coll = io.ImageCollection(str,load_func=img_zoom43)
    
    if(zoomflag == 1):
        coll = io.ImageCollection(str,load_func=img_zoom41)
    else:
        coll = io.ImageCollection(str,load_func=img_zoom42)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\zm_'+ np.str(zoomflag) + '_' + np.str(i)+'.jpg',coll[i])

# ############################
# # gaussian zoom 高斯缩放 图像(金字塔图像)
def img_gaussian_zoom(imgfile):
    img = io.imread(imgfile)
    rows, cols, dim = img.shape # 获取图片的行数 列数和通道数
    # #产生高斯金字塔图像
    # #共生成了log(512)=9幅金字塔图像，加上原始图像共10幅，pyramid[0]-pyramid[1]
    img_pyramid=tuple(transform.pyramid_gaussian(img, downscale=2))
    return img_pyramid[1]

def gaussianzoom_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_gaussian_zoom)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\gsz_' + np.str(i)+'.jpg',coll[i])


# ############################
# # shear 剪切 图像
def ks_shear(x, shear, row_axis=0, col_axis=1, channel_axis=2, fill_mode='nearest', cval=0.):
    shear_matrix = np.array([[1, -np.sin(shear), 0],[0, np.cos(shear), 0],[0, 0, 1]])
    h, w = x.shape[row_axis], x.shape[col_axis]
    transform_matrix = ksimage.transform_matrix_offset_center(shear_matrix, h, w)
    x = ksimage.apply_transform(x, transform_matrix, channel_axis, fill_mode, cval)
    return x

def img_shear41(imgfile):
    img = io.imread(imgfile)
    intensity = 0.2
    sh = np.random.uniform(-intensity, intensity) #逆时针方向剪切强度为正
    img_shear = ks_shear(img, sh)
    return img_shear

def img_shear42(imgfile):
    img = io.imread(imgfile)
    intensity = 0.5
    sh = np.random.uniform(0, intensity) #逆时针方向剪切强度为正
    img_shear = ks_shear(img, sh)
    return img_shear

def img_shear43(imgfile):
    img = io.imread(imgfile)
    intensity = 0.8
    sh = np.random.uniform(0.5, intensity) #逆时针方向剪切强度为正
    img_shear = ks_shear(img, sh)
    return img_shear

# #shearflag = 1 缩0.2 ,2 缩0.5 ,3 缩0.8
def shear_from_dir(imagepath,savepath,shearflag=2):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    if(shearflag == 3):
        coll = io.ImageCollection(str,load_func=img_shear43)
    if(shearflag == 1):
        coll = io.ImageCollection(str,load_func=img_shear41)
    else:
        coll = io.ImageCollection(str,load_func=img_zoom42)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\sr_'+ np.str(shearflag) + '_' + np.str(i)+'.jpg',coll[i])

# ############################
# # contrast 对比度变换 图像增加光照变化
def randomHueSaturationValue(image, hue_shift_limit=(-180, 180),sat_shift_limit=(-255, 255),val_shift_limit=(-255, 255), u=0.5):
    if np.random.random() < u:
        img = color.rgb2hsv(image)
        h, s ,v = img[:,:,0],img[:,:,1],img[:,:,2]
        hue_shift = np.random.uniform(hue_shift_limit[0], hue_shift_limit[1])
        
        h = h + hue_shift
        
        sat_shift = np.random.uniform(sat_shift_limit[0], sat_shift_limit[1])
        s = s + sat_shift
        
        val_shift = np.random.uniform(val_shift_limit[0], val_shift_limit[1])
        v = v + val_shift
        
        img[:,:,0],img[:,:,1],img[:,:,2] = h, s ,v
        
        image = color.hsv2rgb(img)
    return image

def img_contrast(imgfile):
    img = io.imread(imgfile)
    contrast_img = randomHueSaturationValue(img)
    return contrast_img

def contrast_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_contrast)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\ct_' + np.str(i)+'.jpg',coll[i])

# ############################
# # channelshift 随机通道偏移 图像
def random_channel_shift(x, intensity, channel_index=0):
    x = np.rollaxis(x, channel_index, 0)
    min_x, max_x = np.min(x), np.max(x)
    channel_images = [np.clip(x_channel + np.random.uniform(-intensity, intensity), min_x, max_x)
                      for x_channel in x]
    x = np.stack(channel_images, axis=0)
    x = np.rollaxis(x, 0, channel_index+1)
    return x

def img_channelshift(imgfile):
    img = io.imread(imgfile)
    intensity = 0.05
    img_chsh = random_channel_shift(img, intensity)
    return img_chsh

def channelshift_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_channelshift)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\clst_' + np.str(i)+'.jpg',coll[i])


# ############################
# # PCA 白化 图像
def RGB_PCA(images):
    pixels = images.reshape(-1, images.shape[-1])
    idx = np.random.random_integers(0, pixels.shape[0], len(pixels))
    pixels = [pixels[i] for i in idx]
    pixels = np.array(pixels, dtype=np.uint8).T
    m = np.mean(pixels)/256.
    C = np.cov(pixels)/(256.*256.)
    l, v = np.linalg.eig(C)
    return l, v, m

def RGB_variations(image, eig_val, eig_vec):
    a = np.random.randn(3)
    v = np.array([a[0]*eig_val[0], a[1]*eig_val[1], a[2]*eig_val[2]])
    variation = np.dot(eig_vec, v)
    return image + variation

def img_pca(imgfile):
    img = io.imread(imgfile)
    l,v,m = RGB_PCA(img)
    img_pca = RGB_variations(img,l,v)
    return img_pca

def pca_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_pca)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\pca_' + np.str(i)+'.jpg',coll[i])

# ############################
# # 图像调暗
def img_brightness_low(imgfile):
    img = io.imread(imgfile)
    image = img_as_float(img)
    img_gam = exposure.adjust_gamma(image, 2) #调暗
    return img_gam

def lowbrightness_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_brightness_low)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\bnl_' + np.str(i)+'.jpg',coll[i])

# ############################
# # 图像调亮
def img_brightness_high(imgfile):
    img = io.imread(imgfile)
    image = img_as_float(img)
    img_gam = exposure.adjust_gamma(image, 0.5) #调亮
    return img_gam

def highbrightness_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_brightness_high)
    
    # 循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\bnh_' + np.str(i)+'.jpg',coll[i])

# ############################
# # 图像极坐标变换
from PIL import Image
import scipy.misc
import cv2

# ##最邻近插值算法
def interpolate_adjacent(src, ri, rf, rc, ti, tf, tc):
    r = 0
    t = 0
    if (ri - rf) > (rc - ri):
        r = rc
    else:
        r = rf
    if (ti - tf) > (tc - ti):
        t = tc
    else:
        t = tf
    out = src[r][t]
    return out

###双线性插值
def interpolate_bilinear(src, ri, rf, rc, ti, tf, tc):
    if rf == rc & tc == tf:
        out = src[rc][tc]
    elif rf == rc:
        out = (ti - tf)*src[rf][tc] + (tc - ti)*src[rf][tf]
    elif tf == tc:
        out = (ri - rf)*src[rc][tf] + (rc - ri)*src[rf][tf]
    else:
        inter_r1 = (ti - tf)*src[rf][tc] + (tc - ti)*src[rf][tf]
        inter_r2 = (ti - tf)*src[rc][tc] + (tc - ti)*src[rc][tf]
        out = (ri - rf)*inter_r2 + (rc - ri)*inter_r1
    return out  

####三次卷积算法
def interpolate_bitriple(src, ri, rf, rc, ti, tf, tc, rows, cols):
    rf_1 = rf - 1
    rc_1 = rc + 1
    tf_1 = tf - 1
    tc_1 = tc + 1
    
    if rf_1 < 0:
        rf_1 = 0
    if rc_1 > rows - 1:
        rc_1 = rows - 1
    if tf_1 < 0:
        tf_1 = 0
    if tc_1 > cols - 1:
        tc_1 = cols - 1
    
    if rf == rc & tc == tf:
        out = src[rc][tc]
    elif tf == tc:
        A = np.array([(4 - 8*(ri - rf + 1) + 5*np.power(ri - rf + 1, 2) - np.power(ri - rf + 1, 3)),
                      (1 - 2*np.power((ri - rf), 2) + np.power((ri - rf), 3)),
                      (1 - 2*np.power((rc - ri), 2) + np.power((rc - ri), 3)),
                      (4 - 8*(rc - ri + 1) + 5*np.power(rc - ri + 1, 2) - np.power(rc - ri + 1, 3))])
        B = np.array([[src[rf_1][tf]],[src[rf][tf]],[src[rc][tc]],[src[rc_1][tc]]])
        out = np.dot(A, B)
    elif rf == rc:
        B = np.array(src[rf][tf_1],   src[rf][tf],   src[rf][tc],   src[rf][tc_1])
        C = np.array([[(4 - 8*(ti - tf + 1) + 5*np.power(ti - tf + 1, 2) - np.power(ti - tf + 1, 3))],
                      [(1 - 2*np.power((ti - tf), 2) + np.power((ti - tf), 3))],
                      [(1 - 2*np.power((tc - ti), 2) + np.power((tc - ti), 3))],
                      [(4 - 8*(tc - ti + 1) + 5*np.power(tc - ti + 1, 2) - np.power(tc - ti + 1, 3))]])
        out = np.dot(B, C)
    else:  
        A = np.array([(4 - 8*(ri - rf + 1) + 5*np.power(ri - rf + 1, 2) - np.power(ri - rf + 1, 3)),
                      (1 - 2*np.power((ri - rf), 2) + np.power((ri - rf), 3)),
                      (1 - 2*np.power((rc - ri), 2) + np.power((rc - ri), 3)),
                      (4 - 8*(rc - ri + 1) + 5*np.power(rc - ri + 1, 2) - np.power(rc - ri + 1, 3))])

        B = np.array([[src[rf_1][tf_1], src[rf_1][tf], src[rf_1][tc], src[rf_1][tc_1]],
                      [src[rf][tf_1],   src[rf][tf],   src[rf][tc],   src[rf][tc_1]],
                      [src[rc][tf_1],   src[rc][tf],   src[rc][tc],   src[rc][tc_1]],
                      [src[rc_1][tf_1], src[rc_1][tf], src[rc_1][tc], src[rc_1][tc_1]]])

        C = np.array([[(4 - 8*(ti - tf + 1) + 5*np.power(ti - tf + 1, 2) - np.power(ti - tf + 1, 3))],
                      [(1 - 2*np.power((ti - tf), 2) + np.power((ti - tf), 3))],
                      [(1 - 2*np.power((tc - ti), 2) + np.power((tc - ti), 3))],
                      [(4 - 8*(tc - ti + 1) + 5*np.power(tc - ti + 1, 2) - np.power(tc - ti + 1, 3))]])
        out = np.dot(np.dot(A, B), C)
    return out  

# #输入参数src: 源图像矩阵（方图，笛卡尔坐标）
# #输入参数d: 目标图像直径
# #返回：dst：目标图像矩阵（圆图，极坐标）
def polar_transform(src, d):
    # #初始换目标图像矩阵
    dst = np.zeros((d, d), dtype=np.float)
    # #dst = np.zeros((d, d))
    rows, cols, dim = src.shape # 获取图片的行数 列数和通道数
    # rows, cols = src.shape
    
    line_len = rows
    line_num = cols
    
    # #径向步长
    delta_r = line_len/((d - 1)/2.0)
    
    # #切向（旋转角度）步长  
    delta_t = 2.0*np.pi/line_num
    ox = (d - 1)/2.0
    oy = (d - 1)/2.0
    
    for i in range(0, d, 1):
        for j in range(0, d, 1):
            rx = j - ox
            ry = i - oy
            r = np.sqrt(np.power(rx, 2) + np.power(ry, 2))
            
            if (r <= (d - 1)/2.0):              
                t = np.arctan2(ry, rx)
                if t < 0:
                    t = t + 2.0*np.pi
                    
                ri = r*delta_r
                rf = np.int_(np.floor(ri))
                rc = np.int_(np.ceil(ri))
                
                if rf < 0:
                    rf = 0
                if rc > rows - 1:
                    rc = rows - 1
                
                ti = t/delta_t
                tf = np.int_(np.floor(ti))
                tc = np.int_(np.ceil(ti))
                
                if tf < 0:
                    tf = 0
                if tc > cols - 1:
                    tc = cols - 1
                
                # #选择相应的插值算法
                dst[i][j],val,temp = interpolate_adjacent(src, ri, rf, rc, ti, tf, tc)
                # dst[i][j] = interpolate_bilinear(src, ri, rf, rc, ti, tf, tc)
                # dst[i][j] = interpolate_bitriple(src, ri, rf, rc, ti, tf, tc, rows, cols)
    
    return dst

# #输入参数src: 源图像矩阵（方图，笛卡尔坐标）
# #输入参数d: 目标图像直径
# #返回：dst：目标图像矩阵（圆图，极坐标）
def polar3_transform(src, d):
    shape = src.shape
    
    # #初始化 目标图像矩阵
    dst = np.zeros(shape)
    cols = width = shape[1] # 获取图片的列数 也就是宽
    rows = height = shape[0] # 获取图片的行数 也就是高
    # lens = len(shape)
  
    line_len = height # rows
    line_num = width # cols
    
    # #径向步长
    delta_r = line_len/((d - 1)/2.0)
    
    # #切向（旋转角度）步长  
    delta_t = 2.0*np.pi/line_num
    ox = (d - 1)/2.0
    oy = (d - 1)/2.0
    
    for i in range(0, d, 1):
        for j in range(0, d, 1):
            rx = j - ox
            ry = i - oy
            r = np.sqrt(np.power(rx, 2) + np.power(ry, 2))
            
            if (r <= (d - 1)/2.0):              
                t = np.arctan2(ry, rx)
                if t < 0:
                    t = t + 2.0*np.pi
                    
                ri = r*delta_r
                rf = np.int_(np.floor(ri))
                rc = np.int_(np.ceil(ri))
                
                if rf < 0:
                    rf = 0
                if rc > rows - 1:
                    rc = rows - 1
                
                ti = t/delta_t
                tf = np.int_(np.floor(ti))
                tc = np.int_(np.ceil(ti))
                
                if tf < 0:
                    tf = 0
                if tc > cols - 1:
                    tc = cols - 1
                
                # #选择相应的插值算法
                dst[i][j][0],val,temp = interpolate_adjacent(src, ri, rf, rc, ti, tf, tc)
                dst[i][j][1] = val
                dst[i][j][2] = temp
                # interpolate_bilinear(src, ri, rf, rc, ti, tf, tc)
                # dst[i][j][0],val,temp = interpolate_bilinear(src, ri, rf, rc, ti, tf, tc)
                # dst[i][j][1] = val
                # dst[i][j][2] = temp
                # interpolate_bitriple(src, ri, rf, rc, ti, tf, tc, rows, cols)
                # dst[i][j][0],val,temp = interpolate_bitriple(src, ri, rf, rc, ti, tf, tc, rows, cols)
                # dst[i][j][1] = val
                # dst[i][j][2] = temp
    
    return dst

def img_polar(imgfile):
    img = io.imread(imgfile)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    # img_p = polar_transform(img, rows)
    img_p = polar3_transform(img, rows)
    return img_p
    #width = cols
    #height = rows
    #im = Image.new('RGB',(width, height),(255,255,255))
    #for w in range(width):
    #    for h in range(height):
    #        try:
    #            im.putpixel((w,h),tuple(img_p[h][w]))
    #        except:
    #            #im.putpixel((w,h),tuple(img_p[h][w],)*3)
    #            im.putpixel((w,h),255)
    #return im

def polar1_from_dir(imagepath,savepath):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    coll = io.ImageCollection(str,load_func=img_polar)
    # io.imshow(coll[0])
    
    # ok ok
    # scipy.misc.toimage(coll[0], cmin=0.0, cmax=255).save('outfile.jpg')
    # scipy.misc.toimage(coll[0], mode='P').save('outfile.jpg')
    # scipy.misc.toimage(coll[0]).save('outfile.jpg')

    # output = scipy.misc.toimage(coll[0])
    # output.convert('RGBA')
    # scipy.misc.toimage(output).save('outfile.jpg')
   
    # error
    # imgp = Image.fromarray(coll[0])
    # imgp.save('outfile.jpeg')
    
    # ok ok 但是灰色图
    # cv2.imwrite("filename.jpg", coll[0])
    
    # error 
    # matplotlib.image.imsave('name.png', coll[0])
    
    # 循环保存图片
    for i in range(len(coll)):
    #    imgp = Image.fromarray(coll[i])
    #    error
    #    io.imsave(savepath + '\\plr1_' + np.str(i)+'.jpg',coll[i])
        img_name = savepath + '\\plr1_' + np.str(i)+'.jpg'
        scipy.misc.toimage(coll[i]).save(img_name)



# ############################
# #另一种极坐标转换方法
# #
import math

def to_gray(img):
    w, h,_ = img.shape
    ret = np.empty((w, h), dtype=np.uint8)
    retf = np.empty((w, h), dtype=np.float)
    imgf = img.astype(float)
    retf[:, :] = ((imgf[:, :, 1] + imgf[:, :, 2] + imgf[:, :, 0])/3)
    ret = retf.astype(np.uint8)
    return ret

def radia_transform(im,w,h):
    shape = im.shape

    new_im = np.zeros(shape)
    # print(shape)
    # print(len(shape))
    # print('w',w)
    # print('h',h)
    width = shape[1]
    height = shape[0]
    lens = len(shape)
    for i in range(0,width):
        xita = 2*3.14159*i/width
        for a in range(0,height):
            x = (int)(math.floor(a * math.cos(xita)))
            y = (int)(math.floor(a * math.sin(xita)))
            new_y = (int)(h+x)
            new_x = (int)(w+y)
            # print(h.dtype)
            if new_x>=0 and new_x<width:
                if new_y>=0 and new_y<height:
                    if lens==3:
                        new_im[a,i,0] = (im[new_y,new_x,0]-127.5)/128
                        new_im[a,i,1] = (im[new_y,new_x,1]-127.5)/128
                        new_im[a,i,2] = (im[new_y,new_x,2]-127.5)/128
                    else:
                        new_im[a,i] = (im[new_y,new_x]-127.5)/128
                        new_im[a,i] = (im[new_y,new_x]-127.5)/128
                        new_im[a,i] = (im[new_y,new_x]-127.5)/128
    return new_im

def img_polar2_1(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols/10), int(rows))
    # img_p = radia_transform(img, int(cols/10), int(rows/4*3))
    # img_p = radia_transform(img, int(cols/2), int(rows/4*3))
    # img_p = radia_transform(img, int(cols/2), int(rows/2))
    # img_p = radia_transform(img, int(cols/4), int(rows/4))
    # img_p = radia_transform(img, int(cols/4*3), int(rows/4*3))
    # img_p = radia_transform(img, int(cols), int(rows/4*3))
    
    return img_p

def img_polar2_2(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols/10), int(rows/4*3))
    return img_p

def img_polar2_3(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols/2), int(rows/4*3))
    return img_p

def img_polar2_4(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols/2), int(rows/2))
    return img_p

def img_polar2_5(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols/4), int(rows/4))
    return img_p

def img_polar2_6(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols/4*3), int(rows/4*3))
    return img_p

def img_polar2_7(imgfile):
    img = io.imread(imgfile)
    # img = to_gray(img)
    rows, cols, dim = img.shape # 获取图片的行数(高) 列数(宽)和通道数
    img_p = radia_transform(img, int(cols), int(rows/4*3))
    return img_p

def polar2_from_dir(imagepath,savepath,flag=1):
    # str=imagepath+'\\*.jpg'+':'+imagepath+'\\*.png'
    str=imagepath+'\\*.jpg'
    
    if(flag == 1):
        coll = io.ImageCollection(str,load_func=img_polar2_1)
    if(flag == 2):
        coll = io.ImageCollection(str,load_func=img_polar2_2)
    if(flag == 3):
        coll = io.ImageCollection(str,load_func=img_polar2_3)
    if(flag == 4):
        coll = io.ImageCollection(str,load_func=img_polar2_4)
    if(flag == 5):
        coll = io.ImageCollection(str,load_func=img_polar2_5)
    if(flag == 6):
        coll = io.ImageCollection(str,load_func=img_polar2_6)
    if(flag == 7):
        coll = io.ImageCollection(str,load_func=img_polar2_7)
    else:
        coll = io.ImageCollection(str,load_func=img_polar2_1)
    
    # scipy.misc.toimage(coll[0]).save('outfile.jpg')
    # #循环保存图片
    for i in range(len(coll)):
        io.imsave(savepath + '\\plr2_' + np.str(i)+'.jpg',coll[i])


# ############################
# # test
#curpath = os.getcwd()
#print(curpath)
    
# #resize 同比例的大小
# #resize_from_dir('./x-image2','./x-gen')

# #灰化处理
# gray_from_dir('./x-image2','./x-gen')

# #按比例缩放
# rescale_from_dir('./x-image2','./x-gen',1)

# #加噪音
# noise_from_dir('./x-image2','./x-gen')

# #反转 水平 垂直
# flipV_from_dir('./x-image2','./x-gen')
# flipH_from_dir('./x-image2','./x-gen')

# #旋转
# rotate_from_dir('./x-image2','./x-gen',2)

# #平移
# shift_from_dir('./x-image2','./x-gen',2)

# #缩放
# zoom_from_dir('./x-image2','./x-gen',2)
# #高斯缩放
# gaussianzoom_from_dir('./x-image2','./x-gen')

# #对比度变换
# contrast_from_dir('./x-image2','./x-gen')

# #pca白化 error
# pca_from_dir('./x-image2','./x-gen')

# #图像调暗
# lowbrightness_from_dir('./x-image2','./x-gen')
# #图像调亮
# highbrightness_from_dir('./x-image2','./x-gen')

# #图像的极坐标变换
# #polar_from_dir('./x-image2','./x-gen')


# ############################
# # 对x光机图像的增强处理

# #首先resize 同样大小的图像 64x64
# resize_from_dir('./x-image2/pingping','./x-gen')

# #按比例缩放
# rescale_from_dir('./x-gen','./x-pingping-gen',1)
# rescale_from_dir('./x-gen','./x-pingping-gen',2)

# #加噪音
# noise_from_dir('./x-gen','./x-pingping-gen')

# #反转 水平 垂直
# flipV_from_dir('./x-gen','./x-pingping-gen')
# flipH_from_dir('./x-gen','./x-pingping-gen')

# #旋转
# rotate_from_dir('./x-gen','./x-pingping-gen',2)
# rotate_from_dir('./x-gen','./x-pingping-gen',4)
# rotate_from_dir('./x-gen','./x-pingping-gen',6)
# rotate_from_dir('./x-gen','./x-pingping-gen',8)

# #平移
# shift_from_dir('./x-gen','./x-pingping-gen',2)
# shift_from_dir('./x-gen','./x-pingping-gen',3)

# #缩放
# zoom_from_dir('./x-gen','./x-pingping-gen',2)

# #高斯缩放
# gaussianzoom_from_dir('./x-gen','./x-pingping-gen')

# #图像的极坐标变换
polar1_from_dir('./x-gen','./x-pingping-gen')
# #polar2_from_dir('./x-gen','./x-pingping-gen',2)