Contrast Limited Adaptive Histogram Equalization (CLAHE)

enhancing image contrast and brightness through a classical image processing approach.

Works on both RGB and GRAY scale PNG images

Before vs After

Original	CLAHE Enhanced

Features

• CLAHE for grayscale images
• CLAHE for RGB images (applied on each channel separately or with blending)
• Contrast clipping support
• Bilinear interpolation for smooth transition between tiles
• Alpha blending between equalized and original values (for RGB only)
• Png images support implemented using STB image read/write (stb_image, stb_image_write)

How to Use

1. Provide PNG image tou want enhanced: load_png(image path)
2. Call clahe(...) function with the right paramaters — it will enhance the image where you can save to enjoy it using save_png(path, img).

How It Works

Step 1: Tile Division

Divide the image into square tiles of size M × M.

• M is a user-defined parameter (e.g., 64, 128).

---

Step 2: Histogram & Clipping

For each tile and for each color channel:

• Compute the histogram.
• Apply clipping:
  Clip any bin value above a clip limit and accumulate the total excess.
  • Clip limit = (tile width × tile height) / clip factor
  • Clip factor is a user-defined parameter.

---

Step 3: Redistribute Excess

• Distribute the total excess equally across all 256 bins:
  • excess per bin = total excess / 256
• Add this redistributed excess to the original (clipped) histogram bins.

---

Step 4: Cumulative Distribution Function (CDF)

• Calculate the cumulative distribution from the histogram:

  c(0) = H(0) / N  
  c(i) = c(i-1) + H(i) / N
  

  where N = tile width × tile height

---

Step 5: Histogram Equalization

Apply a blend of the equalized value and the original intensity:

HE(i) = alpha * c(i) * i + (1 - alpha) * i

• alpha is the blend parameter in range [0.0, 1.0]
  • 1.0 = full equalization
  • 0.0 = original image

---

Step 6: Interpolation

To avoid visible borders between tiles, interpolate the result for each pixel from the 4 surrounding tiles using bilinear interpolation:

y
↑
|   (0,1)          (1,1)
|    HE10         HE11
|      +------------+
|      |            |
|      |   (fx,fy)  |
|      |            |
|      +------------+
|   (0,0)          (1,0)
|    HE00         HE01
+------------------------→ x

P[i] =
  (1 - fx) * (1 - fy) * HE00[i] +
  fx * (1 - fy) * HE01[i] +
  (1 - fx) * fy * HE10[i] +
  fx * fy * HE11[i];

---

Step 7: Done

Enjoy the adaptively equalized image with improved contrast and brightness :)

---

Parameters to Tune

Parameter	Description
Tile Size M	Size of each square tile
Clip Factor	Lower = more aggressive clipping
Alpha	Blend between original and equalized image

---

Developed by Mohammed Yasser Mohammed

email : es-mohamed.yasser2027@alexu.edu.eg