Basic Image Processing Manipulations

---

Problem Definition

The task is to perform basic image transformations: rotation right and left, flipping, translation from color to grayscale image and smoothing.

---

Method and Implementation

I implemented the functions on Python and used openCV library for image I/O. Color-to-grayscale translation (rgb_to_grayscale) was performed by averaging the color intensities in each pixel (X = (R+G+B)/3). Rotation (transpose_left, transpose_right) and flipping flip_image was done using matrix transposition and flipping. I used box filter for smoothing (box_blur, box_filter). Box filter is a simple Gaussian smoothing interpolation method that assigns average intensity of neighbouring pixels within boxes of a given size for each pixel in the image. I intended to use it to perform detail enhancement via the guided filter [Kaiming He, 2010, ICCV], but due to the time limitation I couldn't make it work as it is supposed to.

---

Results

Results
Input Image	Grayscale	Rotated Left	Rotated Right	Flipped	Blurred

---

Discussion

The box filter that was used for Gaussian smoothing approximation is relatively fast ans simple to implement, however it produces some undesired artifacts (horizontal and vertical lines) since it has wave-like kernel shape in the Fourier domain. Therefore, it is not suitable for high-quality smoothing, and one should use more exact approximations of Gaussian smoothing such as [Yoshizawa and Yokota, IEEE ICIP 2014] instead.

---

---

Credits and Bibliography

• He, Kaiming, Jian Sun, and Xiaoou Tang. "Guided image filtering." European conference on computer vision. Springer, Berlin, Heidelberg, 2010.
• Yoshizawa, Shin, and Hideo Yokota. "Fast L 1 Gaussian convolution via domain splitting." Image Processing (ICIP), 2014 IEEE International Conference on. IEEE, 2014.

---