Image thresholding (binary, adaptive, Otsu) in Computer Vision - Model Metrics & Evaluation

Metrics & Evaluation - Image thresholding (binary, adaptive, Otsu)

Which metric matters for Image Thresholding and WHY

In image thresholding, the goal is to separate objects from the background clearly. The key metric is accuracy of pixel classification: how many pixels are correctly labeled as foreground or background. This is important because a good thresholding method should minimize mistakes in pixel labeling to produce a clean binary image.

For adaptive and Otsu thresholding, metrics like precision and recall on foreground pixels help understand if the method captures the object well without including too much background noise.

Confusion Matrix for Pixel Classification

      | Predicted Background | Predicted Foreground |
      |----------------------|----------------------|
      | True Negative (TN)    | False Positive (FP)   |
      | False Negative (FN)   | True Positive (TP)    |

      Total pixels = TP + FP + TN + FN

      Example:
      TN = 9000 (background pixels correctly identified)
      FP = 500  (background pixels wrongly marked as foreground)
      FN = 300  (foreground pixels missed)
      TP = 1200 (foreground pixels correctly identified)
    

Precision vs Recall Tradeoff in Thresholding

Precision tells us how many pixels labeled as foreground are actually foreground. High precision means less background noise in the object.

Recall tells us how many actual foreground pixels were detected. High recall means the object is fully captured.

For example, in medical imaging, missing parts of a tumor (low recall) is worse than including some background (lower precision). So recall is more important.

In document scanning, including background noise (low precision) can make text unclear, so precision is more important.

Good vs Bad Metric Values for Image Thresholding

• Good: Accuracy above 95%, Precision and Recall both above 90%, meaning most pixels are correctly classified with minimal noise and missing parts.
• Bad: Accuracy below 80%, Precision or Recall below 70%, indicating many pixels are wrongly classified, causing noisy or incomplete binary images.

Common Pitfalls in Thresholding Metrics

• Accuracy paradox: If background pixels dominate, high accuracy can be misleading even if foreground detection is poor.
• Ignoring class imbalance: Foreground pixels are often fewer; metrics like precision and recall are better than accuracy alone.
• Overfitting threshold: Choosing a threshold that works only on one image but fails on others.
• Data leakage: Using test image info to pick threshold biases results.

Self Check: Is 98% Accuracy but 12% Recall Good?

No, this is not good. The model correctly labels most pixels overall, but it only finds 12% of the actual foreground pixels. This means it misses most of the object, which defeats the purpose of thresholding. High recall is crucial to capture the full object.

Key Result

In image thresholding, precision and recall on foreground pixels are key to evaluate how well the object is separated from background, beyond overall accuracy.

Practice

1. What is the main purpose of image thresholding in computer vision?

easy

A. To convert an image into black and white for easier analysis

B. To increase the color depth of an image

C. To blur the image for noise reduction

D. To resize the image to smaller dimensions

Solution

1. Step 1: Understand image thresholding

   Image thresholding simplifies images by turning pixels into black or white based on a cutoff value.

2. Step 2: Identify the purpose

   This simplification helps in easier analysis like object detection or segmentation.

3. Final Answer:

   To convert an image into black and white for easier analysis -> Option A

4. Quick Check:

   Image thresholding = black and white conversion [OK]

Hint: Thresholding means black and white conversion [OK]

Common Mistakes:

• Confusing thresholding with image resizing
• Thinking thresholding increases color depth
• Mixing thresholding with blurring

2. Which of the following is the correct syntax to apply binary thresholding using OpenCV in Python?

easy

A. ret, thresh = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY)

B. ret, thresh = cv2.adaptiveThreshold(image, 127, 255, cv2.THRESH_BINARY)

C. thresh = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY)

D. ret, thresh = cv2.threshold(image, 255, 127, cv2.THRESH_BINARY)

Solution

1. Step 1: Recall OpenCV binary threshold syntax

   The function cv2.threshold returns two values: the threshold used and the thresholded image.

2. Step 2: Check parameter order and function call

   Correct call is cv2.threshold(image, threshold_value, max_value, threshold_type).

3. Final Answer:

   ret, thresh = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY) -> Option A

4. Quick Check:

   cv2.threshold returns two values [OK]

Hint: cv2.threshold returns two values: ret and image [OK]

Common Mistakes:

• Using adaptiveThreshold instead of threshold for binary
• Not unpacking two return values
• Swapping threshold and max values

3. Given the following code snippet, what will be the value of ret after applying Otsu's thresholding?

import cv2
image = cv2.imread('image.jpg', 0)
ret, thresh = cv2.threshold(image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
print(ret)

medium

A. The fixed threshold value 0

B. Always 255

C. The optimal threshold value found by Otsu's method

D. The maximum pixel value in the image

Solution

1. Step 1: Understand Otsu's thresholding output

   When using cv2.THRESH_OTSU, the function ignores the input threshold (0 here) and calculates an optimal threshold automatically.

2. Step 2: Identify what ret holds

   The variable ret stores the threshold value found by Otsu's method, not the input or max pixel value.

3. Final Answer:

   The optimal threshold value found by Otsu's method -> Option C

4. Quick Check:

   Otsu returns optimal threshold in ret [OK]

Hint: Otsu's ret is the best threshold found [OK]

Common Mistakes:

• Assuming ret is always 0 or max pixel value
• Confusing input threshold with output
• Thinking ret is the thresholded image

4. Identify the error in this adaptive thresholding code snippet and select the correct fix:

import cv2
image = cv2.imread('image.jpg', 0)
thresh = cv2.adaptiveThreshold(image, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 6, 2)

medium

A. Image must be read in color mode, not grayscale

B. Max value should be 127 instead of 255

C. Use cv2.THRESH_OTSU instead of cv2.THRESH_BINARY

D. Block size must be an odd number greater than 1; change 6 to 7

Solution

1. Step 1: Check adaptiveThreshold parameters

   The block size parameter must be an odd number greater than 1 to define the neighborhood size.

2. Step 2: Identify the error in block size

   The block size is 6, which is even and will cause a runtime error. It must be changed to an odd number greater than 1, such as 7.

3. Final Answer:

   Block size must be an odd number greater than 1; change 6 to 7 -> Option D

4. Quick Check:

   Block size odd and >1 [OK]

Hint: Block size in adaptiveThreshold must be odd > 1 [OK]

Common Mistakes:

• Using even block size causing runtime error
• Confusing max value with threshold value
• Reading image in color instead of grayscale

5. You have an image with uneven lighting. Which thresholding method should you choose to get the best binary segmentation, and why?

hard

A. Binary thresholding with a fixed value, because it is simple and fast

B. Adaptive thresholding, because it calculates thresholds locally for different regions

C. Otsu's thresholding, because it finds a global optimal threshold automatically

D. No thresholding, just use the original image

Solution

1. Step 1: Understand the problem of uneven lighting

   Uneven lighting means different parts of the image have different brightness levels, making a single global threshold ineffective.

2. Step 2: Compare thresholding methods

   Binary thresholding uses one fixed value, which fails with uneven lighting. Otsu's method finds one global threshold, also insufficient. Adaptive thresholding calculates thresholds for small regions, handling uneven lighting well.

3. Final Answer:

   Adaptive thresholding, because it calculates thresholds locally for different regions -> Option B

4. Quick Check:

   Uneven lighting = adaptive thresholding best [OK]

Hint: Uneven light? Use adaptive thresholding [OK]

Common Mistakes:

• Choosing global threshold methods for uneven lighting
• Ignoring lighting variation in images
• Skipping thresholding and using raw image