OpenCV、Skimage、PIL图像处理的细节差异

上一篇博客中介绍了caffe实现多label输入,其中有一些图片处理的操作(训练验证数据),当时我选择了PIL库,但是在测试中用了caffe官网demo的代码,它使用了caffe提供的一些python接口,而它调用的是skimage这个库,所以有些许差异,可能会带来精度上的一些影响。这篇博客先介绍一下这些库的差异,下篇博客介绍测试时可能出现的小错误

PIL(RGB)

首先介绍PIL(Python Imaging Library)这个库,这是Python中最基础的图像处理库,主要注意对图片进行处理时w,h的变化

from PIL import Image
import numpy as np
image = Image.open('test.jpg') # 图片是400x300 宽x高
print type(image) # out: PIL.JpegImagePlugin.JpegImageFile
print image.size  # out: (400,300)
print image.mode # out: 'RGB'
print image.getpixel((0,0)) # out: (143, 198, 201)
# resize w*h
image = image.resize((200,100),Image.NEAREST)
print image.size # out: (200,100)
'''
代码解释
**注意image是 class:`~PIL.Image.Image` object**,它有很多属性,比如它的size是(w,h),通道是RGB,,他也有很多方法,比如获取getpixel((x,y))某个位置的像素,得到三个通道的值,x最大可取w-1,y最大可取h-1
比如resize方法,可以实现图片的放缩,具体参数如下
resize(self, size, resample=0) method of PIL.Image.Image instance
    Returns a resized copy of this image.

    :param size: The requested size in pixels, as a 2-tuple:
       (width, height). 
    注意size是 (w,h),和原本的(w,h)保持一致
    :param resample: An optional resampling filter.  This can be
       one of :py:attr:`PIL.Image.NEAREST`, :py:attr:`PIL.Image.BOX`,
       :py:attr:`PIL.Image.BILINEAR`, :py:attr:`PIL.Image.HAMMING`,
       :py:attr:`PIL.Image.BICUBIC` or :py:attr:`PIL.Image.LANCZOS`.
       If omitted, or if the image has mode "1" or "P", it is
       set :py:attr:`PIL.Image.NEAREST`.
       See: :ref:`concept-filters`.
    注意这几种插值方法,默认NEAREST最近邻(分割常用),分类常用BILINEAR双线性,BICUBIC立方
    :returns: An :py:class:`~PIL.Image.Image` object.

'''
image = np.array(image,dtype=np.float32) # image = np.array(image)默认是uint8
print image.shape # out: (100, 200, 3)
# 神奇的事情发生了,w和h换了,变成(h,w,c)了
# 注意ndarray中是 行row x 列col x 维度dim 所以行数是高,列数是宽

Skimage(RGB)

skimage即是Scikit-Image,官网

import skimage
from skimage import io,transform
import numpy as np
image= io.imread('test.jpg',as_grey=False)
# 第一个参数是文件名可以是网络地址,第二个参数默认为False,True时为灰度图
print type(image) # out: numpy.ndarray
print image.dtype # out: dtype('uint8')
print image.shape # out: (300, 400, 3) (h,w,c)前面介绍了ndarray的特点
# mode也是RGB
print image
'''
注意此时image里都是整数uint8,范围[0-255]
array([
        [ [143, 198, 201 (dim=3)],[143, 198, 201],... (w=200)],
        [ [143, 198, 201],[143, 198, 201],... ],
        ...(h=100)
      ], dtype=uint8)

'''
image= io.imread('test.jpg',as_grey=True)
print image.shape # out: (300, 400)
print image
'''
此时image范围变为[0-1]
array([[ 0.73148549,  0.73148549,  0.73148549, ...,  0.73148549,
         0.73148549,  0.73148549],
       [ 0.73148549,  0.73148549,  0.73148549, ...,  0.73148549,
       .....]])
'''
print image.dtype # out: dtype('float64')

image = io.imread('test.jpg',as_grey=False) 
# h*w
image = transform.resize(image,(100, 200),order=1) # order默认是1,双线性
#resize后image范围又变成[0-1]
print image.dtype # out: dtype('float64')
print image.shape # out: (100, 200, 3)
print image
'''
array([[[ 0.56078431,  0.77647059,  0.78823529],
        [ 0.56078431,  0.77647059,  0.78823529],
        [ 0.56078431,  0.77647059,  0.78823529],
        ..., ...]])
'''
'''
resize函数接口
resize(image, output_shape, order=1, mode='constant', cval=0, clip=True, preserve_range=False)
order : int, optional
        The order of interpolation. The order has to be in the range 0-5:
         - 0: Nearest-neighbor
         - 1: Bi-linear (default)
         - 2: Bi-quadratic
         - 3: Bi-cubic
         - 4: Bi-quartic
         - 5: Bi-quintic

'''
print skimage.img_as_float(image).dtype # out: float64
# img_as_float可以把image转为double,即float64

OpenCV(python版)(BGR)

OpenCV是个很强大的图像处理库,性能也很好。

import cv2
import numpy as np
image = cv2.imread('test.jpg')
print type(image) # out: numpy.ndarray
print image.dtype # out: dtype('uint8')
print image.shape # out: (300, 400, 3) (h,w,c) 和skimage类似
print image # BGR
'''
array([
        [ [143, 198, 201 (dim=3)],[143, 198, 201],... (w=200)],
        [ [143, 198, 201],[143, 198, 201],... ],
        ...(h=100)
      ], dtype=uint8)

'''
# w*h
image = cv2.resize(image,(100,200),interpolation=cv2.INTER_LINEAR)
print image.dtype # out: dtype('uint8')
print image.shape # out: (200, 100, 3) 
'''
注意注意注意 和skimage不同 
resize(src, dsize[, dst[, fx[, fy[, interpolation]]]]) 
关键字参数为dst,fx,fy,interpolation
dst为缩放后的图像
dsize为(w,h),但是image是(h,w,c)
fx,fy为图像x,y方向的缩放比例,
interplolation为缩放时的插值方式,有三种插值方式:
cv2.INTER_AREA:使用象素关系重采样。当图像缩小时候,该方法可以避免波纹出现。当图像放大时,类似于 CV_INTER_NN方法    
cv2.INTER_CUBIC: 立方插值
cv2.INTER_LINEAR: 双线形插值 
cv2.INTER_NN: 最近邻插值
[详细可查看该博客](http://www.tuicool.com/articles/rq6fIn)
'''
'''
cv2.imread(filename, flags=None):
flag:
cv2.IMREAD_COLOR 1: Loads a color image. Any transparency of image will be neglected. It is the default flag. 正常的3通道图
cv2.IMREAD_GRAYSCALE 0: Loads image in grayscale mode 单通道灰度图
cv2.IMREAD_UNCHANGED -1: Loads image as such including alpha channel 4通道图
注意: 默认应该是cv2.IMREAD_COLOR,如果你cv2.imread('gray.png'),虽然图片是灰度图,但是读入后会是3个通道值一样的3通道图片

'''

在进行图像处理时一点要注意 各个库之间的细微差异,还有要注意图像放缩时插值方法的选择,而且即使是相同的插值方法,各个库的实现也不同,结果也会有些许差异。

阅读更多

扫码向博主提问

爆米花好美啊

非学,无以致疑;非问,无以广识
  • 擅长领域:
  • deep learn
  • detection
  • segmentati
去开通我的Chat快问
版权声明:本文为博主原创文章,如若转载,请注明出处! https://blog.csdn.net/u013010889/article/details/54347089
想对作者说点什么? 我来说一句

没有更多推荐了,返回首页

关闭
关闭
关闭