Python为8bit深度图像应用color map

图片中存在着色版的概念，二维矩阵的每个元素的值指定了一种颜色，因此可以显示出彩色。

迁移调色板

下述python代码将VOC数据集中的某个语义分割的图片的调色板直接应用在一个二维矩阵代表的图像上

#label_im is a numpy array of 1 x height x width
#return an Image object，call its' save('out.png') functioin to save as image file
def palette( label_im):
        import copy
        from PIL import Image
        palette_im = Image.open('2008_000144.png')
        palette = palette_im.palette
        '''
        Transfer the VOC color palette to an output mask for visualization.
        '''
        if label_im.ndim == 3:
                label_im = label_im[0]
                label = Image.fromarray(label_im, mode='P')
                label.palette = copy.copy(palette)
                return label

应用color map

#直接转成含RGB信息的三维矩阵
#示例代码中应用了gist_earth的color map
from matplotlib import cm
im = Image.fromarray(np.uint8(cm.gist_earth(label)*255))

自定义color map

下面的代码用于生成一个color map（由VOC数据集中的代码VOCdevkit/VOCcode/VOClabelcolormap.m转换而来）

import numpy as np
# bitget bitshift bitor zeros is all in matlab internal function
def bitget(num,i):
        ar=np.array([[num]], dtype=np.uint8)
        bits=np.unpackbits(ar, axis=1)[0]
        idx=bits.size - 1 - i
        return bits[idx]

def bitshift(num,i): #left shift，if i <0 ,then same as left_shift(num,-i)
        return np.right_shift(num,i)

def bitor(x,y):
        return np.bitwise_or(x,y)
#N.B. np.zeros default data type is float and usally color map element is float number that less than 1 [(0~255)/255]
def getColorMap(N):
        #default N is 256
        if N==None:
                N=256

        cmap=np.zeros(N*3, dtype=np.uint8).reshape(N,3)
        for i in range(N):
                idx=i
                r=0;g=0;b=0
                for j in range(8):
                        r = bitor(r, bitshift(bitget(idx,0),7 - j));
                        g = bitor(g, bitshift(bitget(idx,1),7 - j));
                        b = bitor(b, bitshift(bitget(idx,2),7 - j));
                        idx = bitshift(idx,-3);
                cmap[i,0]=r; cmap[i,1]=g; cmap[i,2]=b;
        #cmap = cmap / 255
        return cmap
#ar is 2-dim np.ndarray
def toRGBarray(ar,classes):
        cmap=getColorMap(classes)
        rows=ar.shape[0]
        cols=ar.shape[1]
        r=np.zeros(ar.size*3, dtype=np.uint8).reshape(rows,cols,3)
        for i in range(rows):
                for j in range(cols):
                        r[i,j]=cmap[ar[i,j]]

        return r
if __name__ == '__main__':
        cmap=getColorMap(21)
        print cmap

调用方式：

pic_arr=voccm.toRGBarray(label,21)
im = Image.fromarray(pic_arr,mode='RGB')
im.save('out.png')

小结

除了用作常规的图片存储外，通过给二维数组不同元素赋予颜色的方式可以使我们对数据的空间布局分布有感官的认识，类似于热力图可视化的方式。