理解numpy的axis
研究numpy的axis，是因为使用cv2时，图片数据都是以ndarray存储的，然后为了对其中的矩阵数据进行解析、处理（比如局部马赛克），就需要对axis有所了解。
首先看看cv2读取图片信息后是啥东东：
cv2.imread(/path/to/image, flag)
对于flag的取值：
- cv2.IMREAD_COLOR 读取图片的RGB信息，忽略透明度（Alpha）的channel，按照BGR这3个channel的次序返回numpy.ndarray；若调用imread时，没有设置flag的话，flag默认取这个值
- cv2.IMREAD_GRAYSCALE 只返回灰度图像信息，每一个元素的下标可以理解为(y, x)
- cv2.IMREAD_UNCHANGED 按照原图像信息返回。如果为RGB图像，按照BGR这3个channel的次序返回numpy.ndarray；如果为ARGB，按照BGRA这4个channel的次序返回numpy.ndarray
对于ARGB或者RGB，每一个元素的下标可以理解为[y, x, channel]：
- [:, :, 0] Blue channel
- [:, :, 1] Green channel
- [:, :, 2] Red channel
- [:, :, 3] Alpha channel，即 Transparency channel
将BGR转为RGB：
img_rgb = img_bgr[:,:,::-1]
数据参考：
>>> img_bgr = cv2.imread(‘./01.png’)
>>> img_gray = cv2.imread(‘./01.png’, cv2.IMREAD_GRAYSCALE)
>>> img_ori = cv2.imread(‘./01.png’, cv2.IMREAD_UNCHANGED)
>>> img_bgr.shape
(564, 790, 3)
>>> img_gray.shape
(564, 790)
>>> img_ori.shape
(564, 790, 4)
>>> img_bgr
array([[[ 24, 21, 20],
[ 24, 21, 20], [ 24, 21, 20], ..., [ 24, 21, 20], [ 24, 21, 20], [ 24, 21, 20]], [[ 24, 21, 20], [255, 255, 255], [255, 255, 255], ..., [255, 255, 255], [255, 255, 255], [ 24, 21, 20]], [[ 24, 21, 20], [255, 255, 255], [255, 255, 255], ..., [255, 255, 255], [255, 255, 255], [ 24, 21, 20]], ..., [[ 24, 21, 20], [ 69, 41, 34], [ 68, 40, 33], ..., [251, 248, 245], [249, 246, 242], [ 24, 21, 20]], [[ 24, 21, 20], [ 24, 21, 20], [ 24, 21, 20], ..., [ 24, 21, 20], [ 24, 21, 20], [ 24, 21, 20]], [[ 24, 21, 20], [ 24, 21, 20], [ 24, 21, 20], ..., [ 24, 21, 20], [ 24, 21, 20], [ 24, 21, 20]]], dtype=uint8)>>> img_gray
array([[ 21, 21, 21, …, 21, 21, 21],
[ 21, 255, 255, ..., 255, 255, 21], [ 21, 255, 255, ..., 255, 255, 21], ..., [ 21, 43, 43, ..., 248, 245, 21], [ 21, 21, 21, ..., 21, 21, 21], [ 21, 21, 21, ..., 21, 21, 21]], dtype=uint8)>>> img_ori
array([[[ 24, 21, 20, 255],
[ 24, 21, 20, 255], [ 24, 21, 20, 255], ..., [ 24, 21, 20, 255], [ 24, 21, 20, 255], [ 24, 21, 20, 255]], [[ 24, 21, 20, 255], [255, 255, 255, 255], [255, 255, 255, 255], ..., [255, 255, 255, 255], [255, 255, 255, 255], [ 24, 21, 20, 255]], [[ 24, 21, 20, 255], [255, 255, 255, 255], [255, 255, 255, 255], ..., [255, 255, 255, 255], [255, 255, 255, 255], [ 24, 21, 20, 255]], ..., [[ 24, 21, 20, 255], [ 69, 41, 34, 255], [ 68, 40, 33, 255], ..., [251, 248, 245, 255], [249, 246, 242, 255], [ 24, 21, 20, 255]], [[ 24, 21, 20, 255], [ 24, 21, 20, 255], [ 24, 21, 20, 255], ..., [ 24, 21, 20, 255], [ 24, 21, 20, 255], [ 24, 21, 20, 255]], [[ 24, 21, 20, 255], [ 24, 21, 20, 255], [ 24, 21, 20, 255], ..., [ 24, 21, 20, 255], [ 24, 21, 20, 255], [ 24, 21, 20, 255]]], dtype=uint8)
如果暂不看cv2使用cv2.IMREAD_COLOR读取后的数据，单纯看numpy.ndarray，从一维到三维数据，会是这样：
而通过cv2读取的图片数据，每一个元素的下标可以理解为[y, x, channel]，对于flag为cv2.IMREAD_COLOR时，读取后的数据为*(B, G, R)*，每个像素可以通过[y, x, :]进行获取，其axis分析如下：
