Install

pip install imgvision

Author

Xuheng Cao (caoxuhengcn@gmail.com)

How to use?

中文说明-光谱图像评价指标
中文说明-图像颜色空间转换
中文说明-光谱图像可视化

Example

imgvision.spectra()

This Class is to create a standard CIE illuminance for the subsequent operation.

In[1]:  import numpy as np
   import imgvision as iv
   Illuminate_D65 = iv.spectra('d65')
   Illuminate_A = iv.spectra('a')
   Illuminate_D75 = iv.spectra('d75')
   Illuminate_D50 = iv.spectra('d50')

imgvision.spectra.space()

The following example shows a mapping between a hyperspectral image with 31 bands from 400nm~700nm and the corresponding sRGB image.

In[2]:  import matplotlib.pyplot as plt
   sample_SI = np.random.rand(100,100,31)
   sample_MI = Illuminate_D65.space(sample_SI,'srgb')
   plt.imshow(sample_MI)
   plt.show()

downsample

The downsample function aims at downsampling with any scale factor for an image.

In[3]:  LR_sample_SI = iv.downsample(sample_SI,5)
     print(LR_sample_SI.shape)
Out: (20, 20, 31)

imgvision.color.rgb2hsv()

This function is to convert an RGB image (no matter how is it depth) to the HSV color space. hsv2rgb excutes the contrary operation.

In[4]:  sample_HSV = iv.color.rgb2hsv(sample_MI)
In[5]:  sample_RGB = iv.color.hsv2rgb(sample_HSV)

imgvision.distance.cosine()

Calculate the Cosine Similarity distance between two color images and return a 1-D vector, the vector include the Cosine Similarity distance of each element. This function supports the shape of input as a 2-D matrix (an RGB vector) and a 3-D tensor (a color image).

In[6]: a= np.random.rand(128,128,3)
   b= np.random.rand(128,128,3)
   dist = iv.distance.cosine(a,b)
   print(f'Shape: {dist.shape}\nMean distance: {dist.mean()}')
Out: Shape: (16384,)
   Mean distance: 0.2055142334948004

imgvision.cluster.cosine_predict(img,centre)

Predict clusters each pixel of a given image belongs to according to Cosine Similarity distance and given cluster center.

In[7]:  img = np.random.rand(128,128,3)
    centre = np.array([ [0.3,0.5,0.1],
                [0.8,0.1,0.3] ])
    cluster_id = cluster.cosine_predict(img,centre)
    print(cluster_id)
Out: [1 1 1 ... 1 0 1]

imgvision.spectra_metric()

The following example shows a quantitative evaluation between a hyperspectral image with 31 bands from 400nm to 700nm and the corresponding reconstructed image.

In[8]:  sample_GT = np.random.rand(100,100,31)
   sample_SI = np.random.rand(100,100,31)
   Metric = iv.spectra_metric(sample_GT,sample_SI, scale=32)
   PSNR = Metric.PSNR()
   SSIM = Metric.SSIM()
   MSE = Metric.MSE()
   SAM = Metric.SAM()
   ERGAS = Metric.ERGAS()
   CC = Metric.CC()
   print(PSNR,SSIM,MSE,SAM,EGRAS)
Out: [7.765805543824359 0.0041294934819296365 0.16728452073841815 41.049350569568816 5.116354915189658]