epicyon/blurhash.py

"""
Copyright (c) 2019 Lorenz Diener

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

* The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
* You and any organization you work for may not promote white supremacy, hate 
speech and homo- or transphobia - this license is void if you do.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

https://github.com/halcy/blurhash-python

Pure python blurhash decoder with no additional dependencies, for 
both de- and encoding.

Very close port of the original Swift implementation by Dag Ågren.
"""

import math

# Alphabet for base 83
alphabet = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz#$%*+,-.:;=?@[]^_{|}~"
alphabet_values = dict(zip(alphabet, range(len(alphabet))))

def base83_decode(base83_str):
    """
    Decodes a base83 string, as used in blurhash, to an integer.
    """
    value = 0
    for base83_char in base83_str:
        value = value * 83 + alphabet_values[base83_char]
    return value

def base83_encode(value, length):
    """
    Decodes an integer to a base83 string, as used in blurhash.
    
    Length is how long the resulting string should be. Will complain
    if the specified length is too short.
    """
    if int(value) // (83 ** (length)) != 0:
        raise ValueError("Specified length is too short to encode given value.")
        
    result = ""
    for i in range(1, length + 1):
        digit = int(value) // (83 ** (length - i)) % 83
        result += alphabet[int(digit)]
    return result

def srgb_to_linear(value):
    """
    srgb 0-255 integer to linear 0.0-1.0 floating point conversion.
    """
    value = float(value) / 255.0
    if value <= 0.04045:
        return value / 12.92
    return math.pow((value + 0.055) / 1.055, 2.4)

def sign_pow(value, exp):
    """
    Sign-preserving exponentiation.
    """
    return math.copysign(math.pow(abs(value), exp), value)

def linear_to_srgb(value):
    """
    linear 0.0-1.0 floating point to srgb 0-255 integer conversion.
    """
    value = max(0.0, min(1.0, value))
    if value <= 0.0031308:
        return int(value * 12.92 * 255 + 0.5)
    return int((1.055 * math.pow(value, 1 / 2.4) - 0.055) * 255 + 0.5)

def blurhash_components(blurhash):
    """
    Decodes and returns the number of x and y components in the given blurhash.
    """
    if len(blurhash) < 6:
        raise ValueError("BlurHash must be at least 6 characters long.")
    
    # Decode metadata
    size_info = base83_decode(blurhash[0])
    size_y = int(size_info / 9) + 1
    size_x = (size_info % 9) + 1
    
    return size_x, size_y

def blurhash_decode(blurhash, width, height, punch = 1.0, linear = False):
    """
    Decodes the given blurhash to an image of the specified size.
    
    Returns the resulting image a list of lists of 3-value sRGB 8 bit integer
    lists. Set linear to True if you would prefer to get linear floating point 
    RGB back.
    
    The punch parameter can be used to de- or increase the contrast of the
    resulting image.
    
    As per the original implementation it is suggested to only decode
    to a relatively small size and then scale the result up, as it
    basically looks the same anyways.
    """
    if len(blurhash) < 6:
        raise ValueError("BlurHash must be at least 6 characters long.")
    
    # Decode metadata
    size_info = base83_decode(blurhash[0])
    size_y = int(size_info / 9) + 1
    size_x = (size_info % 9) + 1
    
    quant_max_value = base83_decode(blurhash[1])
    real_max_value = (float(quant_max_value + 1) / 166.0) * punch
    
    # Make sure we at least have the right number of characters
    if len(blurhash) != 4 + 2 * size_x * size_y:
        raise ValueError("Invalid BlurHash length.")
        
    # Decode DC component
    dc_value = base83_decode(blurhash[2:6])
    colours = [(
        srgb_to_linear(dc_value >> 16),
        srgb_to_linear((dc_value >> 8) & 255),
        srgb_to_linear(dc_value & 255)
    )]
    
    # Decode AC components
    for component in range(1, size_x * size_y):
        ac_value = base83_decode(blurhash[4+component*2:4+(component+1)*2])
        colours.append((
            sign_pow((float(int(ac_value / (19 * 19))) - 9.0) / 9.0, 2.0) * real_max_value,
            sign_pow((float(int(ac_value / 19) % 19) - 9.0) / 9.0, 2.0) * real_max_value,
            sign_pow((float(ac_value % 19) - 9.0) / 9.0, 2.0) * real_max_value
        ))
    
    # Return image RGB values, as a list of lists of lists, 
    # consumable by something like numpy or PIL.
    pixels = []
    for y in range(height):
        pixel_row = []
        for x in range(width):
            pixel = [0.0, 0.0, 0.0]

            for j in range(size_y):
                for i in range(size_x):
                    basis = math.cos(math.pi * float(x) * float(i) / float(width)) * \
                            math.cos(math.pi * float(y) * float(j) / float(height))
                    colour = colours[i + j * size_x]
                    pixel[0] += colour[0] * basis
                    pixel[1] += colour[1] * basis
                    pixel[2] += colour[2] * basis
            if linear == False:
                pixel_row.append([
                    linear_to_srgb(pixel[0]),
                    linear_to_srgb(pixel[1]),
                    linear_to_srgb(pixel[2]),
                ])
            else:
                pixel_row.append(pixel)
        pixels.append(pixel_row)
    return pixels
    
def blurhash_encode(image, components_x = 4, components_y = 4, linear = False):
    """
    Calculates the blurhash for an image using the given x and y component counts.
    
    Image should be a 3-dimensional array, with the first dimension being y, the second
    being x, and the third being the three rgb components that are assumed to be 0-255 
    srgb integers (incidentally, this is the format you will get from a PIL RGB image).
    
    You can also pass in already linear data - to do this, set linear to True. This is
    useful if you want to encode a version of your image resized to a smaller size (which
    you should ideally do in linear colour).
    """
    if components_x < 1 or components_x > 9 or components_y < 1 or components_y > 9: 
        raise ValueError("x and y component counts must be between 1 and 9 inclusive.")
    height = float(len(image))
    width = float(len(image[0]))
    
    # Convert to linear if neeeded
    image_linear = []
    if linear == False:
        for y in range(int(height)):
            image_linear_line = []
            for x in range(int(width)):
                image_linear_line.append([
                    srgb_to_linear(image[y][x][0]),
                    srgb_to_linear(image[y][x][1]),
                    srgb_to_linear(image[y][x][2])
                ])
            image_linear.append(image_linear_line)
    else:
        image_linear = image
        
    # Calculate components
    components = []
    max_ac_component = 0.0
    for j in range(components_y):
        for i in range(components_x):
            norm_factor = 1.0 if (i == 0 and j == 0) else 2.0
            component = [0.0, 0.0, 0.0]
            for y in range(int(height)):
                for x in range(int(width)):
                    basis = norm_factor * math.cos(math.pi * float(i) * float(x) / width) * \
                                          math.cos(math.pi * float(j) * float(y) / height)
                    component[0] += basis * image_linear[y][x][0]
                    component[1] += basis * image_linear[y][x][1]
                    component[2] += basis * image_linear[y][x][2]
                    
            component[0] /= (width * height)
            component[1] /= (width * height)
            component[2] /= (width * height)
            components.append(component)
            
            if not (i == 0 and j == 0):
                max_ac_component = max(max_ac_component, abs(component[0]), abs(component[1]), abs(component[2]))
                
    # Encode components
    dc_value = (linear_to_srgb(components[0][0]) << 16) + \
               (linear_to_srgb(components[0][1]) << 8) + \
               linear_to_srgb(components[0][2])
    
    quant_max_ac_component = int(max(0, min(82, math.floor(max_ac_component * 166 - 0.5))))
    ac_component_norm_factor = float(quant_max_ac_component + 1) / 166.0
    
    ac_values = []
    for r, g, b in components[1:]:
        ac_values.append(
            int(max(0.0, min(18.0, math.floor(sign_pow(r / ac_component_norm_factor, 0.5) * 9.0 + 9.5)))) * 19 * 19 + \
            int(max(0.0, min(18.0, math.floor(sign_pow(g / ac_component_norm_factor, 0.5) * 9.0 + 9.5)))) * 19 + \
            int(max(0.0, min(18.0, math.floor(sign_pow(b / ac_component_norm_factor, 0.5) * 9.0 + 9.5))))
        )
        
    # Build final blurhash
    blurhash = ""
    blurhash += base83_encode((components_x - 1) + (components_y - 1) * 9, 1)
    blurhash += base83_encode(quant_max_ac_component, 1)
    blurhash += base83_encode(dc_value, 4)
    for ac_value in ac_values:
        blurhash += base83_encode(ac_value, 2)
    
    return blurhash