Deep Learning -- 数据增强
数据增强
在图像的深度学习中,为了丰富图像训练集,更好的提取图像特征,泛化模型(防止模型过拟合),一般都会对数据图像进行数据增强,数据增强,常用的方式,就是旋转图像,剪切图像,改变图像色差,扭曲图像特征,改变图像尺寸大小,增强图像噪音(一般使用高斯噪音)等,但需要注意,不要加入其它图像轮廓的噪音。在不同的任务背景下,我们可以通过图像的几何变换,使用一下一种或者多种组合数据增强变换来增加输入数据的量。
- 旋转|反射变换(Rotation/reflection):随机旋转图像一定角度;改变图像的内容朝向;
- 翻转变换(flip):沿这水平或者垂直方向翻转图像
- 缩放变换(zoom):按照一定的比例放大或者缩小图像
- 平移变换(shift):在图像平面上对图像以一定方式进行平移
数据增强的代码实现
# -*- coding:utf-8 -*- # 数据增强 # 1.翻转变换flip # 2.随机修剪random crop # 3.色彩抖动color jittering # 4.平移变换shift # 5.尺度变换scale # 6.对比度变换contrast # 7.噪声扰动noise # 8.旋转变换/反射变换 Rotation/reflection from PIL import Image,ImageEnhance,ImageOps,ImageFile import numpy as np import random import threading,os,time import logging logger = logging.getLogger(__name__) ImageFile.LOAD_TRUNCATED_IMAGES = True class DataAugmentation: #包含数据增强的八种方式 def __init__(self): pass @staticmethod def openImage(image): return Image.open(image,mode="r") @staticmethod def randomRotation(image,mode=Image.BICUBIC): # 对图像进行任意0~360度旋转 # param mode 邻近插值,双线性插值,双三次B样条插值(default) # param image PIL的图像image # return 旋转之后的图像 random_angle = np.random.randint(1,360) return image.rotate(random_angle,mode) @staticmethod def randomCrop(image): #对图像随意剪切,考虑到图像大小范围(68*68),使用一个一个大于(36*36)的窗口进行截图 #param image:PIL的图像image #return:剪切之后的图像 image_width = image.size[0] image_height = image.size[1] crop_win_size = np.random.randint(40,68) random_region = ((image_width - crop_win_size ) >> 1 , (image_height - crop_win_size) >> 1 ,(image_width + crop_win_size) >> 1 , (image_height + crop_win_size) >> 1) return image.crop(random_region) @staticmethod def randomColor(image): #对图像进行颜色抖动 #param image:PIL的图像image #return:有颜色色差的图像image #随机因子 random_factor = np.random.randint(0, 31) / 10. #调整图像的饱和度 color_image = ImageEnhance.Color(image).enhance(random_factor) #随机因子 random_factor = np.random.randint(10,21) / 10. #调整图像的亮度 brightness_image = ImageEnhance.Brightness(color_image).enhance(random_factor) #随机因子 random_factor = np.random.randint(10,21) / 10. #调整图像的对比度 contrast_image = ImageEnhance.Contrast(brightness_image).enhance(random_factor) #随机因子 random_factor = np.random.randint(0,31) / 10. #调整图像锐度 sharpness_image = ImageEnhance.Sharpness(contrast_image).enhance(random_factor) return sharpness_image @staticmethod def randomGaussian(image,mean=0.2,sigma=0.3): #对图像进行高斯噪声处理 #param image: #return def gaussianNoisy(im,mean=0.2,sigma=0.3): #对图像做高斯噪音处理 # param im:单通道图像 # param mean:偏移量 # param sigma:标准差 #return: for _i in range(len(im)): im[_i] += random.gauss(mean,sigma) return im #将图像转化为数组 img = np.asanyarray(image) #将数组改为读写模式 img.flags.writeable = True width,height = img.shape[:2] #对image的R,G,B三个通道进行分别处理 img_r = gaussianNoisy(img[:,:,0].flatten(), mean, sigma) img_g = gaussianNoisy(img[:,:,1].flatten(), mean, sigma) img_b = gaussianNoisy(img[:,:,2].flatten(), mean, sigma) img[:,:,0] = img_r.reshape([width,height]) img[:,:,1] = img_g.reshape([width,height]) img[:,:,2] = img_b.reshape([width,height]) return Image.fromarray(np.uint8(img)) @staticmethod def saveImage(image,path): image.save(path) def makeDir(path): try: if not os.path.exists(path): if not os.path.isfile(path): os.makdirs(path) return 0 else: return 1 except Exception, e: print str(e) return -1 def imageOps(func_name, image, des_path, file_name, times = 5): funcMap = {"randomRotation": DataAugmentation.randomRotation, "randomCrop":DataAugmentation.randomCrop, "randomColor":DataAugmentation.randomColor, "randomGaussian":DataAugmentation.randomGaussian } if funcMap.get(func_name) is None: logger.error("%s is not exist" , func_name) return -1 for _i in range(0,times,1): new_image = funcMap[func_name](image) DataAugmentation.saveImage(new_image,os.path.join(des_path,func_name + str(_i) + file_name)) opsList = {"randomRotation", "randomCrop", "randomColor", "randomGaussian"} def threadOPS(path,new_path): #多线程处理事务 #param src_path:资源文件 #param des_path:目的地文件 #return: if os.path.isdir(path): img_names = os.listdir(path) else: img_names = [path] for img_name in img_names: print img_name tmp_img_name = os.path.join(path,img_name) print tmp_img_name if os.path.isdir(tmp_img_name): if makeDir(os.path.join(new_path,img_name)) != -1: threadOPS(tmp_img_name,os.path.join(new_path,img_name)) else: print 'create new dir failure' return -1 elif tmp_img_name.split('.')[1] != "DS_Store": image = DataAugmentation.openImage(tmp_img_name) threadImage = [0] * 5 _index = 0 for ops_name in opsList: threadImage[_index] = threading.Thread(target=imageOps,args=(ops_name,image,new_path,img_name)) threadImage[_index].start() _index += 1 time.sleep(0.2) if __name__ == '__main__': threadOPS("C:\Users\Acheron\PycharmProjects\CNN\pic-image\\train\images","C:\Users\Acheron\PycharmProjects\CNN\pic-image\\train\\newimages")
数据增强实验
原始的待进行数据增强的图像:
1.对图像进行颜色抖动
2.对图像进行高斯噪声处理