import cv2
import numpy as np
import matplotlib.pyplot as plt

def gaussian_filter_builtin(image_path, kernel_size=5, sigma=1.0):
    """
    使用OpenCV内置函数实现高斯滤波
    :param image_path: 输入图像路径
    :param kernel_size: 滤波器大小，必须为奇数
    :param sigma: 高斯核的标准差
    :return: 原图和滤波后的图像
    """
    # 读取图像
    img = cv2.imread(image_path)
    
    # 应用高斯滤波
    filtered = cv2.GaussianBlur(img, (kernel_size, kernel_size), sigma)
    
    return img, filtered

def create_gaussian_kernel(size, sigma):
    """
    创建高斯核
    :param size: 核大小，必须为奇数
    :param sigma: 高斯核的标准差
    :return: 高斯核
    """
    kernel = np.zeros((size, size), dtype=np.float32)
    center = size // 2
    
    # 计算高斯核
    for i in range(size):
        for j in range(size):
            x, y = i - center, j - center
            kernel[i, j] = np.exp(-(x**2 + y**2) / (2 * sigma**2))
    
    # 归一化
    kernel = kernel / np.sum(kernel)
    
    return kernel

def manual_gaussian_filter(image_path, kernel_size=5, sigma=1.0):
    """
    手动实现高斯滤波
    :param image_path: 输入图像路径
    :param kernel_size: 滤波器大小，必须为奇数
    :param sigma: 高斯核的标准差
    :return: 高斯滤波后的图像
    """
    # 读取图像
    img = cv2.imread(image_path)
    img_float = img.astype(np.float32)
    height, width, channels = img.shape
    
    # 创建高斯核
    kernel = create_gaussian_kernel(kernel_size, sigma)
    
    # 计算边界填充大小
    pad_size = kernel_size // 2
    
    # 对图像进行边界填充
    padded_img = np.pad(img_float, ((pad_size, pad_size), (pad_size, pad_size), (0, 0)), mode='edge')
    
    # 创建输出图像
    filtered = np.zeros_like(img_float)
    
    # 应用高斯滤波
    for i in range(height):
        for j in range(width):
            for c in range(channels):
                # 提取当前窗口
                window = padded_img[i:i+kernel_size, j:j+kernel_size, c]
                # 计算加权平均值
                filtered[i, j, c] = np.sum(window * kernel)
    
    return img, filtered.astype(np.uint8)

def separable_gaussian_filter(image_path, kernel_size=5, sigma=1.0):
    """
    使用可分离滤波器实现高斯滤波
    :param image_path: 输入图像路径
    :param kernel_size: 滤波器大小，必须为奇数
    :param sigma: 高斯核的标准差
    :return: 原图和滤波后的图像
    """
    # 读取图像
    img = cv2.imread(image_path)
    
    # 创建一维高斯核
    kernel_1d = cv2.getGaussianKernel(kernel_size, sigma)
    
    # 分别在水平和垂直方向应用滤波器
    temp = cv2.filter2D(img, -1, kernel_1d.T)  # 水平方向
    filtered = cv2.filter2D(temp, -1, kernel_1d)  # 垂直方向
    
    return img, filtered

def adaptive_gaussian_filter(image_path, local_variance_threshold=100):
    """
    自适应高斯滤波
    :param image_path: 输入图像路径
    :param local_variance_threshold: 局部方差阈值
    :return: 原图和自适应滤波后的图像
    """
    # 读取图像
    img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE).astype(np.float32)
    height, width = img.shape
    
    # 计算边界填充大小
    kernel_size = 5
    pad_size = kernel_size // 2
    
    # 对图像进行边界填充
    padded_img = np.pad(img, pad_size, mode='edge')
    
    # 创建高斯核
    sigma = 1.0
    kernel = create_gaussian_kernel(kernel_size, sigma)
    
    # 创建输出图像
    filtered = np.zeros_like(img)
    
    # 计算局部统计信息
    for i in range(height):
        for j in range(width):
            # 提取当前窗口
            window = padded_img[i:i+kernel_size, j:j+kernel_size]
            
            # 计算局部方差
            local_variance = np.var(window)
            
            # 如果局部方差大于阈值，说明该区域可能是边缘或纹理区域
            # 此时减少平滑程度
            if local_variance > local_variance_threshold:
                # 保持原始值或轻微平滑
                filtered[i, j] = img[i, j]
            else:
                # 对平坦区域应用高斯滤波
                filtered[i, j] = np.sum(window * kernel)
    
    return cv2.cvtColor(img.astype(np.uint8), cv2.COLOR_GRAY2BGR), cv2.cvtColor(filtered.astype(np.uint8), cv2.COLOR_GRAY2BGR)

def bilateral_vs_gaussian_comparison(image_path):
    """
    比较高斯滤波和双边滤波的效果
    :param image_path: 输入图像路径
    :return: 原图、高斯滤波结果和双边滤波结果
    """
    # 读取图像
    img = cv2.imread(image_path)
    
    # 高斯滤波
    gaussian_result = cv2.GaussianBlur(img, (15, 15), 2)
    
    # 双边滤波
    bilateral_result = cv2.bilateralFilter(img, 15, 80, 80)
    
    return img, gaussian_result, bilateral_result

# 使用示例
if __name__ == "__main__":
    # 注意：需要提供实际的图像路径
    # img, result = gaussian_filter_builtin('image.jpg', 5, 1.0)
    # manual_img, manual_result = manual_gaussian_filter('image.jpg', 5, 1.0)
    # sep_img, sep_result = separable_gaussian_filter('image.jpg', 5, 1.0)
    # adapt_img, adapt_result = adaptive_gaussian_filter('image.jpg', 100)
    # comp_img, gauss_res, bi_res = bilateral_vs_gaussian_comparison('image.jpg')
    pass