2020-11-01

MLX90640的阵列插值处理-多项式插值由32×24像素到512×384像素（附从温度阵列转化为热成像图代码）

由于大四final year project的需求，需要利用mlx90640热成像芯片处理，由于像素过于低，无法利用灰度图进行人脸识别，因此必须扩展像素。这里link 采用了博主的代码，将Pascal版本改编为python版本的代码。

下面是改编的函数的代码，命名为func.py,函数分为：
PolynomialInterpolationArr(SourceDatas,Dir, times) 。 SourceDatas为一个列表，里面是温度矩阵中的一行温度数据，最后返回值为插值结束后的温度数据。具体插值方法见上面链接。

GetPolyData_U(OriginData,times,coes) 会返回一个系数，该系数会被应用在PolynomialInterpolationArr函数当中。

chazhi(temp_matrix)是为了主函数直观清晰而写的一个函数。该函数将温度矩阵拆分为一维列表并根据PolynomialInterpolationArr函数进行插值。

整体思路是先将24×32的温度矩阵在行的方向上插值两次，变为24×512的温度矩阵。之后将24×512的矩阵在列上进行两次插值，变为384*512的温度矩阵.

这里用了MLX90640采集的一次人脸温度数据。txt文本放不上来，有需要的可以把project下载下来测试一下情况。

import numpy as np

def PolynomialInterpolationArr (SourceDatas,Dir, times):
    arrCount=len(SourceDatas)
    Result=[0]*arrCount*4
    if (Dir==0) :
        startIndex = 2
    else :
        startIndex = 1

    for i in range(arrCount):
        Result[startIndex+i*4]=SourceDatas[i]

    coes = [0] * 4 #这里初始化一下，之后应该就每次变化不用归0

    OriginDatas=np.zeros((2,times))

    for i in range(arrCount-times+1) :
        for j in  range(times) :  # 初始化拟合原始数据
            OriginDatas[0,j] = j * 4
            OriginDatas[1,j] = SourceDatas[i + j]
        coes=GetPolyData_U(OriginDatas, times, coes);

        for h in range(1,4) :
            if times >= 2  :
                tempDou = coes[0]+ (h) * coes[1];
            if times >= 3  :
                tempDou = tempDou+ (h) * (h) * coes[2];
            if times >= 4  :
                tempDou = tempDou+ (h) * (h) * (h) * coes[3];

            Result[startIndex + i * 4 + (h)] = tempDou

    # 两端插值，两端插值直接使用线性插值
    OriginDatas = np.zeros((2, 2))
    OriginDatas[0,0] = 0
    OriginDatas[1,0] = SourceDatas[0]
    OriginDatas[0,1] = 4
    OriginDatas[1,1] = SourceDatas[1]

    GetPolyData_U(OriginDatas, 2, coes)

    if Dir==0:
        tempDou = coes[0] + (-1) * coes[1]
        Result[1] = tempDou
        tempDou = coes[0] + (-2) * coes[1]
        Result[0] = tempDou
    else:
        tempDou = coes[0] + (-1) * coes[1]
        Result[0] = tempDou

    # 末端插值
    for i in range(arrCount - times,arrCount - 1):
        for j in range(2):
            OriginDatas[0,j] = j * 4
            OriginDatas[1,j] = SourceDatas[i + j]
        coes=GetPolyData_U(OriginDatas, 2, coes)

        for j in range(1,4):
            tempDou = coes[0] + j * coes[1]
            Result[startIndex + i * 4 + j] = tempDou

    if Dir==0:
        tempDou = coes[0] + (5) * coes[1]
        Result[arrCount * 4 - 1] = tempDou
    else:
        tempDou = coes[0] + (5) * coes[1]
        Result[arrCount * 4 - 2] = tempDou
        tempDou = coes[0] + (6) * coes[1]
        Result[arrCount * 4 - 1] = tempDou

    return Result


def GetPolyData_U(OriginData,times,coes):
    if ((times < 2) or (times > 4)) :
        times=2

    if times==2:
        x1 = OriginData[0,0]
        x2 = OriginData[0,1]
        y1 = OriginData[1,0]
        y2 = OriginData[1,1]
        coes[1] = (y2 - y1) / x2
        coes[0] = y1

    elif times==3:
        x1 = OriginData[0,0]
        x2 = OriginData[0,1]
        x3 = OriginData[0,2]
        y1 = OriginData[1,0]
        y2 = OriginData[1,1]
        y3 = OriginData[1,2]
        coes[2] = ((y3 - y1) * x2 - (y2 - y1) * x3) / (x2 * x3 * x3 - x2 * x2 * x3)
        coes[1] = (y2 - y1) / x2 - coes[2] * x2
        coes[0] = y1

    elif times==4:
        x1 = OriginData[0,0]
        x2 = OriginData[0,1]
        x3 = OriginData[0,2]
        x4 = OriginData[0,3]
        y1 = OriginData[1,0]
        y2 = OriginData[1,1]
        y3 = OriginData[1,2]
        y4 = OriginData[1,3]

        coes[3] = ((y4 - y1) * x2 - (y2 - y1) * x4) / x2 - ((y3 - y1) * x2 - (y2 - y1) * x3) / (
                x2 * x3 * x3 - x2 * x2 * x3) * (x2 * x4 * x4 - x2 * x2 * x4) / x2
        coes[3] = coes[3] / ((x2 * x4 * x4 * x4 - x2 * x2 * x2 * x4) / x2 - (x2 * x3 * x3 * x3 - x2 * x2 * x2 * x3) / (
                x2 * x3 * x3 - x2 * x2 * x3) * (x2 * x4 * x4 - x2 * x2 * x4) / x2)
        coes[2] = ((y3 - y1) * x2 - (y2 - y1) * x3) / (x2 * x3 * x3 - x2 * x2 * x3) - coes[3] * (
                (x2 * x3 * x3 * x3 - x2 * x2 * x2 * x3) / (x2 * x3 * x3 - x2 * x2 * x3))
        coes[1] = ((y2 - y1) - coes[2] * x2 * x2 - coes[3] * x2 * x2 * x2) / x2
        coes[0] = y1
    return coes


def chazhi(temp_matrix):  #temp_matrix是一个24*32的数组
    temp_row = np.zeros((24, 32 * 16))  # 插值之后的数组
    temp_line = np.zeros((32 * 16, 24 * 16))  # temp_row转置之后的数组

    for i in range(24):
        SourceDatas = temp_matrix[i, :].tolist()
        temp_row[i, :] = PolynomialInterpolationArr((PolynomialInterpolationArr(SourceDatas, 0, 4)), 1, 4)
    temp_row = np.transpose(temp_row)  # 方便继续插值

    for i in range(32 * 16):
        SourceDatas = temp_row[i, :].tolist()
        temp_line[i, :] = PolynomialInterpolationArr((PolynomialInterpolationArr(SourceDatas, 0, 4)), 1, 4)

    expandtemp = np.transpose(temp_line)
    return expandtemp
    print(len(expandtemp))

附上main函数代码

import func
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
import matplotlib
from scipy.interpolate import griddata



f = open(r"temp.txt")
val_list = f.readlines()
lists =[]

for string in val_list:
    string = string.split()
    lists.append(string[0:32])
a = np.array(lists)
a= a.astype(float)
#a = np.matrix(a).reshape(24,32)
a=a.reshape(24,32)
#a=np.transpose(a)  #转置，担心a旋转了

expandtemp=func.chazhi(a)



#####
#interpolation=nearest  插值
#origin=upper 镜像
#cmap=plt.cm.      jet\hsv\hot\cool\spring\summer\autumn\winter\gray\bone\copper\pink\lines
#####
#plt.imshow(a,interpolation='nearest',cmap=plt.cm.hot,origin='upper')
plt.imshow(expandtemp,interpolation='None',cmap=plt.cm.gray,origin='upper')

#温标  shrink=0.8
plt.xticks(())
plt.yticks(())
plt.colorbar()

plt.show()
#保存图片
#matplotlib.image.imsave('test.png',a)

图1：插值后根据温度矩阵转化来的灰度图（将plt.imshow中的gray改为hot即可得到热图）

图2：原始灰度图

希望接下来进行的人脸识别能对我友好一点呜呜呜┭┮﹏┭┮。