[DL]DL中常用的图片变换方法

在使用cnn进行物体检测或者物体识别之类任务中，往往都会对原图进行变化处理，已增加数据的复杂度减小overfitting的可能性。下面列出了部分常用的变换方法：

import cv2
import numpy as np
import random
class ImageProcess():
    def RandomBrightness(self,bgr):#
        if random.random() < 0.5:
            hsv = self.BGR2HSV(bgr)
            h,s,v = cv2.split(hsv)
            adjust = random.choice([0.5,1.5])
            v = v*adjust
            v = np.clip(v, 0, 255).astype(hsv.dtype)
            hsv = cv2.merge((h,s,v))
            bgr = self.HSV2BGR(hsv)
        return bgr
    def RandomSaturation(self,bgr):
        if random.random() < 0.5:
            hsv = self.BGR2HSV(bgr)
            h,s,v = cv2.split(hsv)
            adjust = random.choice([0.5,1.5])
            s = s*adjust
            s = np.clip(s, 0, 255).astype(hsv.dtype)
            hsv = cv2.merge((h,s,v))
            bgr = self.HSV2BGR(hsv)
        return bgr
    def RandomHue(self,bgr):
        if random.random() < 0.5:
            hsv = self.BGR2HSV(bgr)
            h,s,v = cv2.split(hsv)
            adjust = random.choice([0.5,1.5])
            h = h*adjust
            h = np.clip(h, 0, 255).astype(hsv.dtype)
            hsv = cv2.merge((h,s,v))
            bgr = self.HSV2BGR(hsv)
        return bgr

    def randomBlur(self,bgr):
        if random.random()<0.5:
            bgr = cv2.blur(bgr,(5,5))
        return bgr

    def randomShift(self,bgr):
        #平移变换
        if random.random() <0.9:
            height,width,c = bgr.shape
            after_shfit_image = np.zeros((height,width,c),dtype=bgr.dtype)
            after_shfit_image[:,:,:] = (104,117,123) #bgr
            shift_x = random.uniform(-width*0.2,width*0.2)
            shift_y = random.uniform(-height*0.2,height*0.2)
            #print(bgr.shape,shift_x,shift_y)
            #原图像的平移
            if shift_x>=0 and shift_y>=0:
                after_shfit_image[int(shift_y):,int(shift_x):,:] = bgr[:height-int(shift_y),:width-int(shift_x),:]
            elif shift_x>=0 and shift_y<0:
                after_shfit_image[:height+int(shift_y),int(shift_x):,:] = bgr[-int(shift_y):,:width-int(shift_x),:]
            elif shift_x <0 and shift_y >=0:
                after_shfit_image[int(shift_y):,:width+int(shift_x),:] = bgr[:height-int(shift_y),-int(shift_x):,:]
            elif shift_x<0 and shift_y<0:
                after_shfit_image[:height+int(shift_y),:width+int(shift_x),:] = bgr[-int(shift_y):,-int(shift_x):,:]
            print(after_shfit_image.shape)
            return after_shfit_image
        return bgr

    def randomScale(self,bgr):
        #固定住高度，以0.6-1.4伸缩宽度，做图像形变
        if random.random() < 0.5:
            scale = random.uniform(0.6,1.4)
            height,width,c = bgr.shape
            bgr = cv2.resize(bgr,(int(width*scale),height))
            return bgr
        return bgr

    def randomCrop(self,bgr):
        if random.random() < 0.9:
            height,width,c = bgr.shape
            h = random.uniform(0.6*height,height)
            w = random.uniform(0.6*width,width)
            x = random.uniform(0,width-w)
            y = random.uniform(0,height-h)
            x,y,h,w = int(x),int(y),int(h),int(w)
            img_croped = bgr[y:y+h,x:x+w,:]
            return img_croped
        return bgr

    def subMean(self,bgr,mean):
        mean = np.array(mean, dtype=np.float32)
        bgr = bgr - mean
        return bgr

    def random_flip(self,im):
        if random.random() < 0.5:
            im_lr = np.fliplr(im).copy()
            h,w,_ = im.shape
            return im_lr
        return im
    def random_bright(self, im, delta=16):
        alpha = random.random()
        if alpha > 0.3:
            im = im * alpha + random.randrange(-delta,delta)
            im = im.clip(min=0,max=255).astype(np.uint8)
        return im
ii=ImageProcess()
img=cv2.imread('1.png')
print(img.shape)
cv2.imshow('1',ii.random_flip(img))
cv2.imshow('2',ii.random_flip(img))
cv2.imshow('3',ii.random_flip(img))
cv2.imshow('4',ii.random_flip(img))
cv2.waitKey(0)