Sample to convert image to Wavelet with Python

I became interested in wavelet transform for a while, so I actually tried to see what it was like.

Things necessary

The following three are required. Let's put it in with pip etc.

• PyWavelets
• numpy
• PIL

Brief commentary

• I am using a library called PyWavelets.
• Discrete wavelet transform (and inverse transform), hierarchical? It does wavelet transform (and inverse transform). I think I can do something else.
• When using 2D data (image), it will not work unless the vertical and horizontal sizes are the same (although the method may just be strange).

Sample code

gist:image_wavelet_transform.py

image_wavelet_transform.py

# coding: utf8
# 2013/2/1 [email protected]
"""Sample script to get an image of the wavelet transform

Require: pip install PyWavelets numpy PIL

Usage: python image_wavelet_transform.py <filename> (<level>:=3) (wavelet:=db1)
"""

import sys
from PIL import Image
import pywt, numpy

filename = sys.argv[1]
LEVEL = len(sys.argv) > 2 and int(sys.argv[2]) or 3
WAVLET = len(sys.argv) > 3 and sys.argv[3] or "db1"


def merge_images(cA, cH_V_D):
    """numpy.4 array(upper left,(Upper right, lower left, lower right))Stick"""
    cH, cV, cD = cH_V_D
    print cA.shape, cH.shape, cV.shape, cD.shape
    cA = cA[0:cH.shape[0], 0:cV.shape[1]] #If the original image is not a power of 2, there may be fractions, so adjust the size. Match the smaller one.
    return numpy.vstack((numpy.hstack((cA,cH)), numpy.hstack((cV, cD)))) #Attach pixels at top left, top right, bottom left, bottom right

def create_image(ary):
    """numpy.Convert array to Grayscale image"""
    newim = Image.new("L", ary.shape)
    newim.putdata(ary.flatten())
    return newim

def wavlet_transform_to_image(gray_image, level, wavlet="db1", mode="sym"):
    """Wavelet transform gray image for level hierarchy and return each stage as image representation

    return [Restore level 0 image,Restore level 1 image,... ,Restore level<level-1>Image of, 各2D係数を１枚Image ofにした画像]
    """
    ret = []
    data = numpy.array(list(gray_image.getdata()), dtype=numpy.float64).reshape(gray_image.size)
    images = pywt.wavedec2(data, wavlet, level=level, mode=mode) # http://www.pybytes.com/pywavelets/ref/2d-dwt-and-idwt.html
    for i in range(2, len(images)+1): #Partially restore and pack in ret
        ary = pywt.waverec2(images[0:i], WAVLET)  * 2**(i-1) / 2**level #The added value is not returned when partially restored(It becomes whitish)So adjust
        ret.append(create_image(ary))
    #Make each 2D coefficient into one image
    merge = images[0] / (2**level) #Values are added to the cA part, so take the average for image display.
    for i in range(1, len(images)): 
        merge = merge_images(merge, images[i]) #Match the four images
    ret.append(create_image(merge))
    return ret

if __name__ == "__main__":
    im = Image.open(filename)
    if im.size[0] != im.size[1]: #If the vertical and horizontal sizes are not the same, something will not work, so I will match it for the time being
        max_size = max(im.size)
        newim = Image.new(im.mode, (max_size, max_size))
        newim.paste(im, (0,0))
        im = newim

    im.getdata() #If you don't call this, for some reason the next split()Fails ...? why?
    bands = im.split() #Process for each RGB channel
    converted_bands_array = [wavlet_transform_to_image(gray, LEVEL, wavlet=WAVLET) for gray in bands] #Convert for each RGB channel
    # zip(*hoge)Is difficult to understand, but converted_For bands(R,G,B)Image of(PIL.Image)Enter.Restore to RGB image with merge
    converted_array = [Image.merge(im.mode, converted_bands) for converted_bands in zip(*converted_bands_array)]

    # converted_array: [Restore level 0 image,Restore level 1 image,... ,Restore level<level-1>Image of, 各2D係数を１枚Image ofにした画像]
    for i, img in enumerate(converted_array): 
        img.save("%s_%d.png " % (filename, i)) #Appropriate image output

Execution result example

Original image (kuma_005.jpg)

Image of coefficient imaged (level = 5) (kuma_005.jpg_5.png)