What is the equivalent of Matlab's imadjust in python?

Question

Is there an equivalent of imadjust in python. equalizeHist does not give a similar result.

Answer 1

You can find the C++ version of the imadjust here: Is there any function equivalent to Matlab's imadjust in OpenCV with C++?

And this version of python code from @maslovw is very good.I just optimize a few loop too make it run a little faster.

import numpy as np
import bisect
from numba import jit

@jit
def imadjust(src, tol=1, vin=[0,255], vout=(0,255)):
    # src : input one-layer image (numpy array)
    # tol : tolerance, from 0 to 100.
    # vin  : src image bounds
    # vout : dst image bounds
    # return : output img

    assert len(src.shape) == 2 ,'Input image should be 2-dims'

    tol = max(0, min(100, tol))

    if tol > 0:
        # Compute in and out limits
        # Histogram
        hist = np.histogram(src,bins=list(range(256)),range=(0,255))[0]

        # Cumulative histogram
        cum = hist.copy()
        for i in range(1, 256): cum[i] = cum[i - 1] + hist[i]

        # Compute bounds
        total = src.shape[0] * src.shape[1]
        low_bound = total * tol / 100
        upp_bound = total * (100 - tol) / 100
        vin[0] = bisect.bisect_left(cum, low_bound)
        vin[1] = bisect.bisect_left(cum, upp_bound)

    # Stretching
    scale = (vout[1] - vout[0]) / (vin[1] - vin[0])
    vs = src-vin[0]
    vs[src<vin[0]]=0
    vd = vs*scale+0.5 + vout[0]
    vd[vd>vout[1]] = vout[1]
    dst = vd

    return dst

Answer 2

One solution is the following piece of code:

def imadjust(x,a,b,c,d,gamma=1):
    # Similar to imadjust in MATLAB.
    # Converts an image range from [a,b] to [c,d].
    # The Equation of a line can be used for this transformation:
    #   y=((d-c)/(b-a))*(x-a)+c
    # However, it is better to use a more generalized equation:
    #   y=((x-a)/(b-a))^gamma*(d-c)+c
    # If gamma is equal to 1, then the line equation is used.
    # When gamma is not equal to 1, then the transformation is not linear.

    y = (((x - a) / (b - a)) ** gamma) * (d - c) + c
    return y

Usage Example:

Matplotlib.pyplot's imshow function requires that the range of the input image is in [0,1]. The following example shows how to read a RGB or gray-scale image, scale the image, and display it.

from PIL import Image
import matplotlib.pyplot as plt
import numpy as np

image = Image.open(fname)
arr = np.asarray(image)
arr2=imadjust(arr,arr.min(),arr.max(),0,1)

fig = plt.figure()
fig.suptitle('image')
plt.imshow(arr2)
plt.show()

Answer 3

Taken from this solution

import numpy as np
import bisect

def imadjust(src, tol=1, vin=[0,255], vout=(0,255)):
    # src : input one-layer image (numpy array)
    # tol : tolerance, from 0 to 100.
    # vin  : src image bounds
    # vout : dst image bounds
    # return : output img

    dst = src.copy()
    tol = max(0, min(100, tol))

    if tol > 0:
        # Compute in and out limits
        # Histogram
        hist = np.zeros(256, dtype=np.int)
        for r in range(src.shape[0]):
            for c in range(src.shape[1]):
                hist[src[r,c]] += 1
        # Cumulative histogram
        cum = hist.copy()
        for i in range(1, len(hist)):
            cum[i] = cum[i - 1] + hist[i]

        # Compute bounds
        total = src.shape[0] * src.shape[1]
        low_bound = total * tol / 100
        upp_bound = total * (100 - tol) / 100
        vin[0] = bisect.bisect_left(cum, low_bound)
        vin[1] = bisect.bisect_left(cum, upp_bound)

    # Stretching
    scale = (vout[1] - vout[0]) / (vin[1] - vin[0])
    for r in range(dst.shape[0]):
        for c in range(dst.shape[1]):
            vs = max(src[r,c] - vin[0], 0)
            vd = min(int(vs * scale + 0.5) + vout[0], vout[1])
            dst[r,c] = vd
    return dst

if you don't want to set vin and vou. Just use cv2.equalizeHist.

@jit
def hisEqul(img):
    return cv2.equalizeHist(img)

@jit
def hisEqulColor(img):
    ycrcb = cv2.cvtColor(img, cv2.COLOR_BGR2YCR_CB)
    channels = cv2.split(ycrcb)
    cv2.equalizeHist(channels[0], channels[0])
    cv2.merge(channels, ycrcb)
    cv2.cvtColor(ycrcb, cv2.COLOR_YCR_CB2BGR, img)
    return img