使用DPC_PINTO算法进行坏点校正

于 2024-07-23 01:54:57 发布
阅读量1k 收藏 18
点赞数 33
分类专栏： ISP图像处理_python 文章标签： 计算机视觉 人工智能 python 算法
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.youkuaiyun.com/2401_83885036/article/details/140622980
版权

DPC_PINTO算法

DPC_PINTO算法通过检查坏点的邻近像素来校正坏点。如果像素被判定为坏点，则用其周围像素的中值替换。该算法主要包含以下步骤：

  1. 读取原始图像：从文件中加载原始拜耳图像。
  2. 扩展图像：扩展图像以处理边界条件。
  3. 检测和校正坏点：使用邻近像素检测和校正坏点。
  4. 显示结果：可视化原始图像和校正后的图像

    实现

    步骤1：读取原始图像

    首先，我们定义一个函数从文件中读取原始图像数据。图像以二进制格式存储，我们将其重塑为正确的维度。

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

def read_raw(file_path, bayer_bits, row, col):
    with open(file_path, 'rb') as f:
        raw_data = np.fromfile(f, dtype=np.uint8 if bayer_bits == 8 else np.uint16)
    raw_data = raw_data.reshape((row, col))
    return raw_data
    步骤2：扩展图像

    接下来，我们扩展图像以便于处理边界条件。这确保了算法在处理边缘像素时不会遇到索引问题

    def expand_raw(img, expand_num):
    if expand_num % 2 != 0:
        raise ValueError('expand_num must be an even number!')
    
    height, width = img.shape
    img_expand = np.zeros((height + expand_num * 2, width + expand_num * 2), dtype=img.dtype)
    img_expand[expand_num:height + expand_num, expand_num:width + expand_num] = img
    
    img_expand[:expand_num, expand_num:width + expand_num] = img[:expand_num, :]
    img_expand[height + expand_num:, expand_num:width + expand_num] = img[-expand_num:, :]
    img_expand[:, :expand_num] = img_expand[:, expand_num:expand_num * 2]
    img_expand[:, width + expand_num:] = img_expand[:, width:width + expand_num]
    
    return img_expand
    步骤3：检测和校正坏点

    该算法的主要功能是检测和校正坏点。judge_defect_pixel函数检查周围像素并确定一个像素是否有缺陷。

    def judge_defect_pixel(around_p, current_p, th):
    median_v = np.median(around_p)
    diff = around_p - current_p
    if (np.all(diff > 0) or np.all(diff < 0)) and np.all(np.abs(diff) > th):
        return median_v
    else:
        return current_p
    步骤4：应用DPC_PINTO算法

    我们将DPC_PINTO算法应用于图像的一个子集。这涉及迭代像素并根据需要校正它们。

    # --------全局变量---------
expand_num = 2
th = 2

# --------原始参数-------
file_path = '../images/HisiRAW_4208x3120_8bits_RGGB.raw'
bayer_format = 'RGGB'
bayer_bits = 8
row = 4208
col = 3120
# ----------------------

start_time = time.time()

raw_data = read_raw(file_path, bayer_bits, row, col)
height, width = raw_data.shape

# 从原始图像中截取一个小块
sub_row_start = 1000
sub_row_end = 1600
sub_col_start = 1000
sub_col_end = 1600
raw_data_sub = raw_data[sub_row_start:sub_row_end, sub_col_start:sub_col_end]

height_sub, width_sub = raw_data_sub.shape
img_expand = expand_raw(raw_data_sub, expand_num)

dis_img = np.zeros((height_sub, width_sub), dtype=raw_data_sub.dtype)

for i in range(expand_num, height_sub + expand_num, 2):
    for j in range(expand_num, width_sub + expand_num, 2):
        # 处理红色像素 (R)
        around_r_pixel = [img_expand[i - 2, j - 2], img_expand[i - 2, j], img_expand[i - 2, j + 2],
                          img_expand[i, j - 2], img_expand[i, j + 2], img_expand[i + 2, j - 2], img_expand[i + 2, j],
                          img_expand[i + 2, j + 2]]
        dis_img[i - expand_num, j - expand_num] = judge_defect_pixel(around_r_pixel, img_expand[i, j], th)

        # 处理绿色像素 (Gr)
        around_gr_pixel = [img_expand[i - 1, j], img_expand[i - 2, j + 1], img_expand[i - 1, j + 2],
                           img_expand[i, j - 1], img_expand[i, j + 3], img_expand[i + 1, j], img_expand[i + 2, j + 1],
                           img_expand[i + 1, j + 2]]
        dis_img[i - expand_num, j - expand_num + 1] = judge_defect_pixel(around_gr_pixel, img_expand[i, j + 1], th)

        # 处理蓝色像素 (B)
        around_b_pixel = [img_expand[i - 1, j - 1], img_expand[i - 1, j + 1], img_expand[i - 1, j + 3],
                          img_expand[i + 1, j - 1], img_expand[i + 1, j + 3], img_expand[i + 3, j - 1],
                          img_expand[i + 3, j + 1], img_expand[i + 3, j + 3]]
        dis_img[i - expand_num + 1, j - expand_num + 1] = judge_defect_pixel(around_b_pixel, img_expand[i + 1, j + 1],
                                                                             th)

        # 处理绿色像素 (Gb)
        around_gb_pixel = [img_expand[i, j - 1], img_expand[i - 1, j], img_expand[i, j + 1], img_expand[i + 1, j - 2],
                           img_expand[i + 1, j + 2], img_expand[i + 2, j - 1], img_expand[i + 3, j],
                           img_expand[i + 2, j + 1]]
        dis_img[i - expand_num + 1, j - expand_num] = judge_defect_pixel(around_gb_pixel, img_expand[i + 1, j], th)

    完整代码

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


def read_raw(file_path, bayer_bits, row, col):
    with open(file_path, 'rb') as f:
        raw_data = np.fromfile(f, dtype=np.uint8 if bayer_bits == 8 else np.uint16)
    raw_data = raw_data.reshape((row, col))
    return raw_data


def expand_raw(img, expand_num):
    if expand_num % 2 != 0:
        raise ValueError('expand_num must be an even number!')

    height, width = img.shape
    img_expand = np.zeros((height + expand_num * 2, width + expand_num * 2), dtype=img.dtype)
    img_expand[expand_num:height + expand_num, expand_num:width + expand_num] = img

    img_expand[:expand_num, expand_num:width + expand_num] = img[:expand_num, :]
    img_expand[height + expand_num:, expand_num:width + expand_num] = img[-expand_num:, :]
    img_expand[:, :expand_num] = img_expand[:, expand_num:expand_num * 2]
    img_expand[:, width + expand_num:] = img_expand[:, width:width + expand_num]

    return img_expand


def judge_defect_pixel(around_p, current_p, th):
    median_v = np.median(around_p)
    diff = around_p - current_p
    if (np.all(diff > 0) or np.all(diff < 0)) and np.all(np.abs(diff) > th):
        return median_v
    else:
        return current_p


# --------全局变量---------
expand_num = 2
th = 2

# --------原始参数-------
file_path = '../images/HisiRAW_4208x3120_8bits_RGGB.raw'
bayer_format = 'RGGB'
bayer_bits = 8
row = 4208
col = 3120
# ----------------------

start_time = time.time()

raw_data = read_raw(file_path, bayer_bits, row, col)
height, width = raw_data.shape

# 从原始图像中截取一个小块
sub_row_start = 1000
sub_row_end = 1600
sub_col_start = 1000
sub_col_end = 1600
raw_data_sub = raw_data[sub_row_start:sub_row_end, sub_col_start:sub_col_end]

height_sub, width_sub = raw_data_sub.shape
img_expand = expand_raw(raw_data_sub, expand_num)

dis_img = np.zeros((height_sub, width_sub), dtype=raw_data_sub.dtype)

for i in range(expand_num, height_sub + expand_num, 2):
    for j in range(expand_num, width_sub + expand_num, 2):
        # 处理红色像素 (R)
        around_r_pixel = [img_expand[i - 2, j - 2], img_expand[i - 2, j], img_expand[i - 2, j + 2],
                          img_expand[i, j - 2], img_expand[i, j + 2], img_expand[i + 2, j - 2], img_expand[i + 2, j],
                          img_expand[i + 2, j + 2]]
        dis_img[i - expand_num, j - expand_num] = judge_defect_pixel(around_r_pixel, img_expand[i, j], th)

        # 处理绿色像素 (Gr)
        around_gr_pixel = [img_expand[i - 1, j], img_expand[i - 2, j + 1], img_expand[i - 1, j + 2],
                           img_expand[i, j - 1], img_expand[i, j + 3], img_expand[i + 1, j], img_expand[i + 2, j + 1],
                           img_expand[i + 1, j + 2]]
        dis_img[i - expand_num, j - expand_num + 1] = judge_defect_pixel(around_gr_pixel, img_expand[i, j + 1], th)

        # 处理蓝色像素 (B)
        around_b_pixel = [img_expand[i - 1, j - 1], img_expand[i - 1, j + 1], img_expand[i - 1, j + 3],
                          img_expand[i + 1, j - 1], img_expand[i + 1, j + 3], img_expand[i + 3, j - 1],
                          img_expand[i + 3, j + 1], img_expand[i + 3, j + 3]]
        dis_img[i - expand_num + 1, j - expand_num + 1] = judge_defect_pixel(around_b_pixel, img_expand[i + 1, j + 1],
                                                                             th)

        # 处理绿色像素 (Gb)
        around_gb_pixel = [img_expand[i, j - 1], img_expand[i - 1, j], img_expand[i, j + 1], img_expand[i + 1, j - 2],
                           img_expand[i + 1, j + 2], img_expand[i + 2, j - 1], img_expand[i + 3, j],
                           img_expand[i + 2, j + 1]]
        dis_img[i - expand_num + 1, j - expand_num] = judge_defect_pixel(around_gb_pixel, img_expand[i + 1, j], th)

# 创建一个包含两个子图的大图
fig, axes = plt.subplots(1, 2, figsize=(12, 6))

# 显示原始图像块
axes[0].imshow(raw_data_sub, cmap='gray')
axes[0].set_title('Original Sub Image')

# 显示校正后的图像块
axes[1].imshow(dis_img, cmap='gray')
axes[1].set_title('Corrected Sub Image')

# 显示处理时间
print('Cost time: {:.2f} seconds'.format(time.time() - start_time))
plt.show()

结果展示 ：

评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值