python实现图像渲染

在本学期结尾，潇小白与小鉴，小强，小鸣，小灿，小哲六个人一起组成了一个项目组花费了一周的时间做成了这个项目，在这里与大家分享。

-- coding: utf-8 --

@Time : 2018/5/22 14:31

@Author : Sun

@File : Final.py

from future import print_function
import time
import numpy as np
from keras.preprocessing.image import load_img, img_to_array
from scipy.misc import imsave
from scipy.optimize import fmin_l_bfgs_b
from keras.applications import vgg16
from keras import backend as K

def Change(base_image_path,style_reference_image_path,result_prefix,iterations):
# 不同损耗成分的权重
total_variation_weight = 1.0
style_weight = 1.0
content_weight = 0.025

# 计算求出迭代得到的图片的长和宽
width, height = load_img(base_image_path).size
img_nrows = 400
img_ncols = int(width * img_nrows / height)

# 调整图片大小并将其格式化成合适的张量
def preprocess_image(image_path):
    img = load_img(image_path, target_size=(img_nrows, img_ncols))
    img = img_to_array(img)
    img = np.expand_dims(img, axis=0)
    img = vgg16.preprocess_input(img)  # 通过vgg16获取数据集特征
    return img

# 将张量转换为有效图像
def deprocess_image(x):
    # channels_first 维度顺序
    if K.image_data_format() == 'channels_first':
        x = x.reshape((3, img_nrows, img_ncols))
        x = x.transpose((1, 2, 0))
    # channels_last 维度顺序
    else:
        x = x.reshape((img_nrows, img_ncols, 3))

    # 用平均像素去除零中心
    x[:, :, 0] += 103.939
    x[:, :, 1] += 116.779
    x[:, :, 2] += 123.68

    # BGR 模式 -> RGB 模式
    x = x[:, :, ::-1]
    x = np.clip(x, 0, 255).astype('uint8')
    return x

# 得到源图像的张量表示
base_image = K.variable(preprocess_image(base_image_path))  # 源图像的张量表示
style_reference_image = K.variable(preprocess_image(style_reference_image_path))  # 风格图像的张量表示

# 迭代图像
# channels_first 维度顺序方式
if K.image_data_format() == 'channels_first':
    combination_image = K.placeholder((1, 3, img_nrows, img_ncols))
# channels_last 维度顺序方式
else:
    combination_image = K.placeholder((1, img_nrows, img_ncols, 3))

# 将三个图像合并成一个Keras张量
input_tensor = K.concatenate([base_image, style_reference_image, combination_image], axis=0)

# 以base_image, style_reference_image, combination_image为输入构建VGG16网络
model = vgg16.VGG16(input_tensor=input_tensor,weights='imagenet', include_top=False)  # 模型加载预先训练的图像权重
print('模型已加载')

outputs_dict = dict([(layer.name, layer.output) for layer in model.layers])  # 获取每个“键”层的符号输出

# 图像张量的克矩阵
def gram_matrix(x):
    assert K.ndim(x) == 3

    # channels_first 维度顺序
    if K.image_data_format() == 'channels_first':
        features = K.batch_flatten(x)
    # channels_last 维度顺序
    else:
        features = K.batch_flatten(K.permute_dimensions(x, (2, 0, 1)))
    gram = K.dot(features, K.transpose(features))
    return gram

# 风格损失
# 基于从样式引用图像和生成的图像中获取特征映射的gram矩阵
def style_loss(style, combination):
    assert K.ndim(style) == 3  # 断言确保数据的正确性
    assert K.ndim(combination) == 3  # 断言确保数据的正确性
    S = gram_matrix(style)
    C = gram_matrix(combination)
    channels = 3  # 通道数
    size = img_nrows * img_ncols
    return K.sum(K.square(S - C)) / (4. * (channels ** 2) * (size ** 2))

# 内容损失
# 用于维护生成图像中基本图像的“内容
def content_loss(base, combination):
    return K.sum(K.square(combination - base))

# 总变异损失
# 保持生成的图像局部一致
def total_variation_loss(x):
    assert K.ndim(x) == 4

    # channels_first 维度顺序
    if K.image_data_format() == 'channels_first':
        a = K.square(x[:, :, :img_nrows - 1, :img_ncols - 1] - x[:, :, 1:, :img_ncols - 1])
        b = K.square(x[:, :, :img_nrows - 1, :img_ncols - 1] - x[:, :, :img_nrows - 1, 1:])
    # channels_last 维度顺序
    else:
        a = K.square(x[:, :img_nrows - 1, :img_ncols - 1, :] - x[:, 1:, :img_ncols - 1, :])
        b = K.square(x[:, :img_nrows - 1, :img_ncols - 1, :] - x[:, :img_nrows - 1, 1:, :])
    return K.sum(K.pow(a + b, 1.25))

# 将上述三个损失函数的结果合并为单个标量
loss = K.variable(0.)
layer_features = outputs_dict['block4_conv2']
base_image_features = layer_features[0, :, :, :]
combination_features = layer_features[2, :, :, :]
loss += content_weight * content_loss(base_image_features,combination_features)

feature_layers = ['block1_conv1', 'block2_conv1','block3_conv1', 'block4_conv1','block5_conv1']
for layer_name in feature_layers:
    layer_features = outputs_dict[layer_name]
    style_reference_features = layer_features[1, :, :, :]
    combination_features = layer_features[2, :, :, :]
    sl = style_loss(style_reference_features, combination_features)
    loss += (style_weight / len(feature_layers)) * sl
loss += total_variation_weight * total_variation_loss(combination_image)

# 获取生成图像的梯度损失
grads = K.gradients(loss, combination_image)

outputs = [loss]
if isinstance(grads, (list, tuple)):
    outputs += grads
else:
    outputs.append(grads)

f_outputs = K.function([combination_image], outputs)

def eval_loss_and_grads(x):
    # channels_first 维度顺序
    if K.image_data_format() == 'channels_first':
        x = x.reshape((1, 3, img_nrows, img_ncols))
    # channels_last 维度顺序
    else:
        x = x.reshape((1, img_nrows, img_ncols, 3))

    outs = f_outputs([x])
    loss_value = outs[0]
    if len(outs[1:]) == 1:
        grad_values = outs[1].flatten().astype('float64')
    else:
        grad_values = np.array(outs[1:]).flatten().astype('float64')
    return loss_value, grad_values

# 求值类，计算传递损失和梯度
class Evaluator(object):
    def __init__(self):
        self.loss_value = None
        self.grads_values = None

    def loss(self, x):
        assert self.loss_value is None
        loss_value, grad_values = eval_loss_and_grads(x)
        self.loss_value = loss_value
        self.grad_values = grad_values
        return self.loss_value

    def grads(self, x):
        assert self.loss_value is not None
        grad_values = np.copy(self.grad_values)
        self.loss_value = None
        self.grad_values = None
        return grad_values

evaluator = Evaluator()  # 新建对象

# 在生成图像的像素上运行基于SCIPI的优化（L-BFGS）,以减少损失
# channels_first 维度顺序
if K.image_data_format() == 'channels_first':
    x = np.random.uniform(0, 255, (1, 3, img_nrows, img_ncols)) - 128.
# channels_last 维度顺序
else:
    x = np.random.uniform(0, 255, (1, img_nrows, img_ncols, 3)) - 128.

for i in range(iterations):
    print('开始第%d次迭代'  %(i+1))
    start_time = time.time()
    x, min_val, info = fmin_l_bfgs_b(evaluator.loss, x.flatten(),fprime=evaluator.grads, maxfun=20)
    img = deprocess_image(x.copy())
    fname = result_prefix + '_at_iteration_%d.png' % (i+1)

    # 保存最后一次迭代得到的图像
    if(i+1 == iterations):
        imsave(fname, img)

    end_time = time.time()
    print('Iteration %d completed in %ds' % (i+1, end_time - start_time))

• from tkinter import *
  from tkinter import ttk
  from PIL import Image,ImageTk
  import win32ui
  from Final import Change

‘’‘图形化界面’’’
class Window(Frame):
def init(self, master=None):
Frame.init(self, master)
self.master = master
self.init_window()

def init_window(self):
    self.master.title("图像风格渲染器")
    self.pack(fill=BOTH, expand=1)
    # 实例化一个Menu对象，这个在主窗体添加一个菜单
    menu = Menu(self.master)
    self.master.config(menu=menu)

    # 创建“文件”菜单，下面有“保存”和“退出”两个子菜单
    file = Menu(menu)
    file.add_command(label='保存')
    file.add_command(label='退出', command=self.client_exit)
    menu.add_cascade(label='文件', menu=file)

    # 创建“帮助”菜单，下面有“使用帮助”和“版权”两个子菜单
    forhelp = Menu(menu)
    forhelp.add_command(label='使用帮助')
    forhelp.add_command(label='版权')
    menu.add_cascade(label='帮助', menu=forhelp)

    frm = Frame(self)
    frm_L = Frame(frm)
    frm_L.grid(row=0,column=10)
    frm_C = Frame(frm)
    frm_C.grid(row=0,column=20)
    frm_R = Frame(frm)
    frm_R.grid(row=0,column=30)

    canvas = Canvas(frm_L, width=400, height=400)
    canvas.pack(padx=5, pady=5)
    canvas.create_rectangle(30, 50, 350, 311)

    canvas = Canvas(frm_R, width=400, height=400)
    canvas.pack(padx=5, pady=5)
    canvas.create_rectangle(30, 50, 350, 311)  #（30,50）为正方形右上角坐标，（350,350）为正方形右下角坐标

    # 标签
    l_empty2 = Label(frm_C, height=2, ).grid(row=95, column=100)  # 空的，用来调位置
    l_times = Label(frm_C, text="迭 代 次 数：", height=2, ).grid(row=10, column=100)
    l_style = Label(frm_C, text="风 格 选 择：", height=2, ).grid(row=50, column=100)

    # 下拉列表
    self.comboxlist_times = ttk.Combobox(frm_C, textvariable=1, height=3, width=10)  # 初始化
    self.comboxlist_times["values"] = ("10", "20", "30", "40")
    self.comboxlist_times.grid(row=20, column=100)
    self.comboxlist_style = ttk.Combobox(frm_C, textvariable="梵高", height=3, width=10)  # 初始化
    self.comboxlist_style["values"] = ("梵高-星空", '梵高-播种者', "毕加索-斗牛", '毕加索-格尔尼卡', '米勒-拾穗')
    self.comboxlist_style.grid(row=60, column=100)

    # 按钮
    b_loader = Button(frm_C, text="加载图片", width=8,command=self.showImg).grid(row=100, column=100)  # 加载按钮
    b_trans = Button(frm_C, text="转换画风", width=8,command=self.catch).grid(row=5, column=100) # 转化按钮
    frm.pack()

# 退出
def client_exit(self):
    exit()

def showImg(self):
    # 1表示打开文件对话框
    dlg = win32ui.CreateFileDialog(1)
    dlg.DoModal()
    # 获取选择的文件名称
    self.name = dlg.GetPathName()
    # 以一个PIL图像对象打开  【调整待转图片格式】
    pil_image = Image.open(self.name)
    # 缩放图像让它保持比例，同时限制在一个矩形框范围内  【调用函数，返回整改后的图片】
    pil_image_resized = resize(325, 325, pil_image)
    # 把PIL图像对象转变为Tkinter的PhotoImage对象  【转换格式，方便在窗口展示】
    tk_image = ImageTk.PhotoImage(pil_image_resized)

    img = Label(self, image=tk_image,width=300,height=250)
    img.image = tk_image
    img.pack()
    img.place(x=45, y=60)

def catch(self):
    base ='./' + self.comboxlist_style.get() +'.jpg'
    Change.Change(self.name, base, './Result', (int)(self.comboxlist_times.get()))
    result_name = './Result_at_iteration_%d.png' % ((int)(self.comboxlist_times.get()))
    pil_image = Image.open(result_name)
    # 缩放图像
    pil_image_resized = resize(325, 325, pil_image)
    # 把PIL图像对象转变为Tkinter的PhotoImage对象
    tk_image = ImageTk.PhotoImage(pil_image_resized)

    img = Label(self, image=tk_image,width=300,height=250)
    img.image = tk_image
    img.pack()
    img.place(x=550, y=60)

参数是：要适应的窗口宽、高、Image.open后的图片

def resize(w_box, h_box, pil_image):
# 获取图像的原始大小
w, h = pil_image.size
f1 = 1.0 * w_box / w
f2 = 1.0 * h_box / h
factor = min([f1, f2])
width = int(w * factor)
height = int(h * factor)
return pil_image.resize((width, height), Image.ANTIALIAS)

root=Tk()

设置界面大小

root.geometry(‘890x400’)

是否可以调节

root.resizable(width=False, height=False)
app = Window(root)
root.mainloop()