计算机视觉的训练任务通常需要很多图像数据来做数据集。常见大数据集如ImageNet, VOC,COCO,等等。我们要完成自己特定场合的识任务,就需要收集或采集场景数据,涉及到一些基本的图像处理方法。我遇到或需要使用到的图像处理方法大概有以下几种:
1.视频切图片(video_to_image.py
)
import cv2
vc = cv2.VideoCapture("/home/ba/Videos/video.mp4") # 读入视频文件,命名cv
n = 1 # 计数
if vc.isOpened(): # 判断是否正常打开
rval, frame = vc.read()
else:
rval = False
timeF = 1 # 视频帧计数间隔频率
i = 0
while rval: # 循环读取视频帧
rval, frame = vc.read()
if (n % timeF == 0): # 每隔timeF帧进行存储操作
i += 1
print(i)
cv2.imwrite("/home/bova/Desktop/open/video/{}.jpg".format(i),
frame) # 存储为图像
n = n + 1
cv2.waitKey(1)
vc.release()
2.视频旋转(video_rotate.py)
有时候,遇到视频是歪的,需要将视频旋转到正常视角,然后再使用1中的方法。这里涉及到opencv的transpose函数
import numpy as np
import cv2
cap = cv2.VideoCapture("/home/bova/Videos/demo.mp4")
fps=25
size=(1080,1920)
# Define the codec and create VideoWriter object
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
out = cv2.VideoWriter('/home/bova/Videos/output.mp4',fourcc, fps, size)
while(cap.isOpened()):
ret, frame = cap.read()
if ret==True:
frame = cv2.transpose(cv2.flip(frame,0)) #0是顺时针90度,-1是逆时针90度
# write the flipped frame
out.write(frame)
cv2.imshow('frame',frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
else:
break
# Release everything if job is finished
cap.release()
out.release()
cv2.destroyAllWindows()
3.视频旋转切图片(video_to_image_rotate.py
)
如果不需要纠正视频,可以直接将视频旋转切片,其实是1和2的组合。
import cv2
vc = cv2.VideoCapture("/home/bova/Desktop/demo.mp4") # 读入视频文件,命名cv
n = 1 # 计数
if vc.isOpened(): # 判断是否正常打开
rval, frame = vc.read()
else:
rval = False
timeF =1 # 视频帧计数间隔频率
i = 0
while rval: # 循环读取视频帧
rval, frame = vc.read()
if (n % timeF == 0): # 每隔timeF帧进行存储操作
i += 1
print(i)
cv2.imwrite("/home/bova/Desktop/demo/{}.jpg".format(i),
cv2.transpose(cv2.flip(frame,0))) # 存储为图像,0顺时针90, -1 逆时针90
n = n + 1
cv2.waitKey(1)
vc.release()
4.按帧数截取一段视频(pull_video.py
)
有时候需要截取视频的某一片段
import cv2
print(cv2.__version__)
videoCapture = cv2.VideoCapture('test.mp4')
fps = 25 #保存视频的帧率
size = (1280,720) #保存视频的大小
videoWriter =cv2.VideoWriter('video4.avi',cv2.VideoWriter_fourcc('X','V','I','D'),fps,size)
i = 0
while True:
success,frame = videoCapture.read()
if success:
i += 1
print('i = ',i)
if(i>=15 and i < 100): #拉取15帧到100帧的视频
videoWriter.write(frame)
else:
print('end')
break
5.图片旋转(image_rotate.py)
from PIL import Image
import os
# 获得文件夹下所有文件
filePath = './img/'
filenames = os.listdir(filePath)
# 指定保存的文件夹
outputPath = './image_rotate/'
# 迭代所有图片
for filename in filenames:
# 读取图像
im = Image.open(filePath + filename)
# 指定逆时针旋转的角度
im_rotate = im.rotate(90)
# 保存图像
im_rotate.save(outputPath + filename)
6.图片合成视频(images_to_video.py
)
如果图片的尺寸一样,可以直接合并,但是如果图片大小不一致,需要先resize到相同大小,再合并。
def resize(img_array, align_mode):
_height = len(img_array[0])
_width = len(img_array[0][0])
for i in range(1, len(img_array)):
img = img_array[i]
height = len(img)
width = len(img[0])
if align_mode == 'smallest':
if height < _height:
_height = height
if width < _width:
_width = width
else:
if height > _height:
_height = height
if width > _width:
_width = width
for i in range(0, len(img_array)):
img1 = cv2.resize(img_array[i], (_width, _height), interpolation=cv2.INTER_CUBIC)
img_array[i] = img1
return img_array, (_width, _height)
"""
如果图片尺寸大小一致,可以直接使用下面的代码:
import cv2
img = cv2.imread('/home/ba/Desktop/demo/1.jpg') #图片所在路径,这里选取第一张获取图片信息
# 设置每秒读取多少张图片
fps = 25
imgInfo = img.shape
# 获取图片宽高度信息
size = (imgInfo[1], imgInfo[0])
videowriter = cv2.VideoWriter("a.mp4",cv2.VideoWriter_fourcc(*"mp4v"),fps,size)
for i in range(1,1700):#图片的总数,最好将图片统一命名为
img = cv2.imread("{}.jpg".format(i))
videowriter.write(img)
print("end")