图像亮度不均场景下Blob检测优化方案技术问询

我有一张亮度分布不均的图像，需要检测其中所有圆点，尝试多种方法后仍存在漏检，具体过程如下：

• 先将图像转为灰度图，应用THRESH_BINARY_INV+THRESH_OTSU得到二值图像，经连通域分析滤除小部件后，即使调参cv2.simpleBlobDetector()仍无法检测所有圆点。
• 尝试先提亮图像再转灰度，微调阈值得到二值图像后，输入LoG、DoG和DoH检测器检测，其中Determinant of Hessian (DoH)表现最优，但仍存在漏检。
• 后续尝试调整DoH的sigma值至0.65、转换至LAB颜色空间对B通道做阈值处理后用轮廓检测、将LAB处理后的二值图输入DoH检测器等方法，依旧无法实现全检测，恳请提供可行优化思路。

已尝试的代码实现 #

基础Blob检测算法代码 #

# Blob detection algorithms
blobs_log = blob_log(binary_image, max_sigma=20, num_sigma=10, threshold=.05)
blobs_log[:, 2] = skimage.feature.blobs_log[:, 2] * np.sqrt(2)

blobs_dog = blob_dog(binary_image, max_sigma=20, threshold=.1)
blobs_dog[:, 2] = skimage.feature.blobs_dog[:, 2] * np.sqrt(2)

blobs_doh = skimage.feature.blob_doh(binary_image, max_sigma=20, threshold=.03)

调整Sigma参数的DoH代码 #

blob_doh(binary_image,min_sigma=0.65,threshold=.0325)

LAB颜色空间处理代码 #

# Convert the image to LAB color space
lab_image = cv2.cvtColor(img, cv2.COLOR_BGR2LAB)

# Extract the B channel (representing the blue-yellow axis)
b_channel = lab_image[:, :, 2]

# Define a threshold value (adjust as needed)
threshold_value = 75  # You can adjust this threshold value

# Apply thresholding on the B channel to make blobs darker than the background
binary_image = cv2.threshold(b_channel, threshold_value, 255, cv2.THRESH_BINARY_INV)[1]

cv2_imshow(binary_image)

轮廓检测代码 #

#Find contours in the binary image
contours, _ = cv2.findContours(binary_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

# Create a copy of the original image to draw circles
result_image = img.copy()

# Draw red circles around the detected dots on the result image
for contour in contours:
    # Calculate the center and radius of the minimum enclosing circle
    (x, y), radius = cv2.minEnclosingCircle(contour)
    center = (int(x), int(y))
    radius = int(radius)
    
    # Draw a red circle with a thickness of 2
    cv2.circle(result_image, center, radius, (0, 0, 255), 1)

# Display the result
cv2_imshow(result_image)

LAB图结合DoH检测代码 #

from skimage.feature import blob_doh
blobs = blob_doh(binary_image,min_sigma=0.65,threshold=.0325)

# Scale the blob coordinates to match the original image size
blobs[:, 0] *= img.shape[1] / binary_image.shape[1]
blobs[:, 1] *= img.shape[0] / binary_image.shape[0]

# Draw circles around the detected blobs on the result image
for blob in blobs:
    y, x, r = blob
    cv2.circle(img, (int(x), int(y)), int(r), (0, 0, 255), 1)  # Draw a red circle with a thickness of 2

# Display the result image with detected blobs
cv2_imshow(img)

可行优化思路建议 #

• 自适应阈值替代全局阈值：使用cv2.adaptiveThreshold()处理灰度图或LAB通道，针对局部亮度不均动态调整阈值，避免全局阈值导致的部分区域圆点信息丢失。
• 对比度增强预处理：对图像先应用CLAHE（限制对比度自适应直方图均衡）增强局部对比度，弱化亮度不均影响后再做二值化，示例代码：

  clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8,8))
  enhanced_gray = clahe.apply(gray_image)
  

• DoH参数网格搜索：固定阈值范围，对min_sigma、max_sigma、num_sigma做网格遍历，找到适配所有圆点的参数组合，避免手动调参的局限性。
• 轮廓筛选规则优化：轮廓检测时，增加圆度（轮廓面积与外接圆面积的比值）、面积范围的筛选条件，过滤非圆点轮廓，同时保留小尺寸或低对比度的圆点轮廓。
• 多检测结果融合：将DoH、轮廓检测、simpleBlobDetector的检测结果做融合，用非极大值抑制（NMS）去除重复检测，合并不同方法的结果以减少漏检。
• 背景建模分离前景：使用背景减法（如cv2.createBackgroundSubtractorMOG2()）提取前景圆点，先分离出目标区域再做检测，规避复杂背景的亮度干扰。

内容的提问来源于stack exchange，提问作者Priyanka Rajendran