zoukankan      html  css  js  c++  java
  • Python: PS 图像特效 — 抽象画风

    今天介绍一种基于图像分割和color map 随机采样生成一种抽象画风的图像特效,简单来说,就是先生成一张 color map 图,颜色是渐变的,然后针对要处理的图像,进行分割,这里用的是 SLIC 分割算法,然后从 color map 中随机采样,将采样得到的像素值赋予分割后的图像区域。

    # -*- coding: utf-8 -*-
    """
    Created on Sun Aug 20 08:31:04 2017
    
    @author: shiyi
    """
    
    import numpy as np
    import matplotlib.pyplot as plt
    from skimage import io
    from skimage.segmentation import slic
    import numpy.matlib
    import random
    
    file_name='D:/Visual Effects/PS Algorithm/9.jpg';
    img=io.imread(file_name)
    
    row, col, channel = img.shape
    
    # define the colormap
    
    color_map = img.copy()
    
    rNW = 0.5
    rNE = 1.0     
    rSW = 0.0
    rSE = 0.5     
    
    gNW = 0.0
    gNE = 0.5
    gSW = 0.0
    gSE = 1.0
    
    bNW = 1.0
    bNE = 0.0
    bSW = 0.5
    bSE = 0.0
    
    xx = np.arange (col) 
    yy = np.arange (row)
    
    x_mask = numpy.matlib.repmat (xx, row, 1)
    y_mask = numpy.matlib.repmat (yy, col, 1)
    y_mask = np.transpose(y_mask)
    
    fx = x_mask * 1.0 / col
    fy = y_mask * 1.0 / row
    
    p = rNW + (rNE - rNW) * fx
    q = rSW + (rSE - rSW) * fx
    r = ( p + (q - p) * fy )
    r[r<0] = 0
    r[r>1] =1
    
    p = gNW + (gNE - gNW) * fx
    q = gSW + (gSE - gSW) * fx
    g = ( p + (q - p) * fy )
    g[g<0] = 0
    g[g>1] =1
    
    p = bNW + (bNE - bNW) * fx
    q = bSW + (bSE - bSW) * fx
    b = ( p + (q - p) * fy )
    b[b<0] = 0.0
    b[b>1] = 1.0
    
    color_map[:, :, 0] = r * 255
    color_map[:, :, 1] = g * 255
    color_map[:, :, 2] = b * 255
    
    # segment the image
    
    N_block = 100
    segments = slic(img, n_segments=N_block, compactness=10)
    # plt.imshow(segments, plt.cm.gray)
    
    seg_img = img.copy()
    T_mask = img.copy()
    
    for i in range(N_block):
        mask = (segments == i)
        T_mask[:, :, 0] = mask
        T_mask[:, :, 1] = mask
        T_mask[:, :, 2] = mask
        x_ind = int(random.random() * (col-1))
        y_ind = int(random.random() * (row-1))
        color = color_map[y_ind, x_ind, :]
        T_img = seg_img * T_mask 
        T_img = color
        seg_img = seg_img * (1-T_mask) + T_img * T_mask
    
    plt.figure(2)
    plt.imshow(seg_img)
    plt.show()
    

    原图:
    这里写图片描述

    效果图:
    这里写图片描述

  • 相关阅读:
    DOSD用scratch的方式训练通用目标检测,性能很高
    caffemodel模型
    NetScope脱机(localhost)使用[转】
    class前置声明
    const函数
    CUDA开发
    caffe2学习
    faster rcnn讲解很细
    控制台输出覆盖当前行显示
    UA池 代理IP池 scrapy的下载中间件
  • 原文地址:https://www.cnblogs.com/mtcnn/p/9412397.html
Copyright © 2011-2022 走看看