图形处理软件中的风格滤镜：从原理到应用的深度解析

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.applications import VGG19
from tensorflow.keras.models import Model

# 加载预训练的 VGG19 模型
vgg = VGG19(include_top=False, weights='imagenet')

# 定义内容层和风格层
content_layer = 'block5_conv2'
style_layers = [
    'block1_conv1',
    'block2_conv1',
    'block3_conv1',
    'block4_conv1',
    'block5_conv1'
]

# 提取特征的模型
def build_model():
    outputs = [vgg.get_layer(name).output for name in style_layers + [content_layer]]
    return Model(inputs=vgg.input, outputs=outputs)

# 计算 Gram 矩阵
def gram_matrix(tensor):
    channels = int(tensor.shape[-1])
    a = tf.reshape(tensor, [-1, channels])
    n = tf.shape(a)[0]
    gram = tf.matmul(a, a, transpose_a=True) / tf.cast(n, tf.float32)
    return gram

# 定义内容和风格损失
def compute_loss(model, content_image, style_image, generated_image):
    outputs = model(tf.concat([style_image, content_image, generated_image], axis=0))
    style_outputs = outputs[:len(style_layers)]
    content_output = outputs[len(style_layers):]

    style_targets = [gram_matrix(style) for style in style_outputs[:len(style_layers)]]
    content_target = content_output[-1]

    style_weight = 1e-2
    content_weight = 1e4

    style_loss = tf.add_n([
        tf.reduce_mean((gram_matrix(gen) - target) ** 2)
        for gen, target in zip(style_outputs[len(style_layers):], style_targets)
    ])

    content_loss = tf.reduce_mean((content_output[-1] - content_target) ** 2)

    loss = style_weight * style_loss + content_weight * content_loss
    return loss

# 优化器
optimizer = tf.optimizers.Adam(learning_rate=0.02)

# 风格迁移主函数
def style_transfer(content_path, style_path, iterations=1000):
    content_image = load_and_process_image(content_path)
    style_image = load_and_process_image(style_path)
    generated_image = tf.Variable(content_image)

    model = build_model()

    for i in range(iterations):
        with tf.GradientTape() as tape:
            loss = compute_loss(model, content_image, style_image, generated_image)

        grads = tape.gradient(loss, generated_image)
        optimizer.apply_gradients([(grads, generated_image)])

        if i % 100 == 0:
            print(f"Iteration {i}: Loss = {loss.numpy()}")

    return generated_image

# 加载和预处理图像
def load_and_process_image(path):
    img = tf.keras.preprocessing.image.load_img(path, target_size=(224, 224))
    img = tf.keras.preprocessing.image.img_to_array(img)
    img = tf.keras.applications.vgg19.preprocess_input(img)
    return tf.convert_to_tensor(np.expand_dims(img, axis=0))

# 示例调用
result = style_transfer('content.jpg', 'style.jpg')
plt.imshow(result[0].numpy())
plt.show()