Image display error after changing dtype of image matrix

Question

I'm using opencv + python to process fundus(retinal images). There is a problem that im facing while converting a float64 image to uint8 image.

Following is the python code:

import cv2
import matplotlib.pyplot as plt
import numpy as np
from tkFileDialog import askopenfilename

filename = askopenfilename()
a = cv2.imread(filename)
height, width, channel = a.shape
b, Ago, Aro = cv2.split(a)
mr = np.average(Aro)
sr = np.std(Aro)

Ar = Aro - np.mean(Aro)
Ar = Ar - mr - sr

Ag = Ago - np.mean(Ago)
Ag = Ag - mr - sr

#Values of elements in Ar
#    Ar = [[-179.17305527, -169.17305527, -176.17305527, ..., -177.17305527, -177.17305527, -177.17305527],
#           [-178.17305527, -169.17305527, -172.17305527, ..., -177.17305527, -177.17305527, -177.17305527],
#           [-179.17305527, -178.17305527, -179.17305527, ..., -177.17305527, -177.17305527, -177.17305527],
#           ...,
#           [-177.17305527, -177.17305527, -177.17305527, ..., -177.17305527, -177.17305527, -177.17305527],
#           [-177.17305527, -177.17305527, -177.17305527, ..., -177.17305527, -177.17305527, -177.17305527],
#           [-177.17305527, -177.17305527, -177.17305527, ..., -177.17305527, -177.17305527, -177.17305527]]

Mr = np.mean(Ar)
SDr = np.std(Ar)

print "MR = ", Mr, "SDr = ", SDr

Mg = np.mean(Ag)
SDg = np.std(Ag)
Thg = np.mean(Ag) + 2 * np.std(Ag) + 50 + 12

Thr = 50 - 12 - np.std(Ar)
print "Thr = ", Thr
Dd = np.zeros((height, width))
Dc = Dd
for i in range(height):
    for j in range(width):

        if Ar[i][j] > Thr:
            Dd[i][j] = 255
        else:
            Dd[i][j] = 0

TDd = np.uint8(Dd)
TDd2 = Dd

for i in range(height):
    for j in range(width):
        if Ag[i][j] > Thg:
            Dc[i][j] = 1
        else:
            Dc[i][j] = 0

#CALCULATING RATIO
ratio = 500.0 / Dd.shape[1]
dim = (500, int(Dd.shape[0] * ratio))
#
# #RESIZING TO-BE-DISPLAYED IMAGES
resized_TDd = cv2.resize(TDd, dim, interpolation=cv2.INTER_AREA)
resized_TDd2 = cv2.resize(TDd2, dim, interpolation=cv2.INTER_AREA)
resized_original = cv2.resize(Aro, dim, interpolation=cv2.INTER_AREA)

cv2.imshow('TDd', resized_TDd)
cv2.imshow('TDd2', resized_TDd2)
cv2.imshow('Aro', resized_original)
cv2.waitKey(0)

Ar[][] has -ve as well as +ve values and Thr has a -ve value. Dd is the image which i want to display. The problem is that TDd displays a bizarre image (255s and 0s are being assigned to appropriate pixels, i checked but the image being displayed is weird and not similar to TDd

Original image

Red channel image:

TDd (uint8 of Dd):

TDd2 (same as Dd)

Dd2 (declared uint8 dtype while initializing)

Why are the TDd and TDd2 images different? Since the difference between the gray values of the pixels (as far as i understand and know) in these 2 images is only 255, 0 (in TDd) and 255.0, 0.0 (in TDd2).

It would be a great great help if someone could tell me.

Luca Angioloni Luca Angioloni · Accepted Answer · 2016-10-15T22:03:57

Usually when Images are represented with np float64 arrays, RGB values are in range 0 to 1. So when converting to uint8 there is a possible precision loss when converting from float64 to uint8.

I would directly create Dd as a:

Dd = np.zeros((height, width), dtype=np.uint8)

Then it should work.

Image display error after changing dtype of image matrix

2 Answers