数据的标准化,归一化方法
数据的标准化(normalization)是将数据按比例缩放,使之落入一个小的特定区间。在某些比较和评价的指标处理中经常会用到,去除数据的单位限制,将其转化为无量纲的纯数值,便于不同单位或量级的指标能够进行比较和加权。
像素值在0-10000之间,通过归一化到 [-1,1] 之间。
import os
import SimpleITK as sitk
def adaptive_normal(image_path, outpath):
"""
normalize image, rescaled to [-1, 1], without background voxels in statistic analysis
:param image_path: image path, nii.gz
:param outpath: .nii.gz, 归一化后的文件保存路径
:return: 归一化后的 nii.gz 文件绝对路径
"""
min_p = 0.001
max_p = 0.999 # quantile prefer 98~99
image = sitk.ReadImage(image_path)
image_array = sitk.GetArrayFromImage(image)
imgArray = np.float32(image_array)
imgPixel = imgArray[imgArray >= 0]
imgPixel.sort()
index = int(round(len(imgPixel) - 1) * min_p + 0.5)
if index < 0:
index = 0
if index > (len(imgPixel) - 1):
index = len(imgPixel) - 1
value_min = imgPixel[index]
index = int(round(len(imgPixel) - 1) * max_p + 0.5)
if index < 0:
index = 0
if index > (len(imgPixel) - 1):
index = len(imgPixel) - 1
value_max = imgPixel[index]
mean = (value_max + value_min) / 2.0
stddev = (value_max - value_min) / 2.0
imgArray = (imgArray - mean) / stddev
imgArray[imgArray < -1] = -1.0
imgArray[imgArray > 1] = 1.0
img = sitk.GetImageFromArray(imgArray, isVector=False)
sitk.WriteImage(img, outpath)
return os.path.abspath(outpath)
https://github.com/Wddzht/DataPreprocessing
Game Over~~~