To change the body type of one person to body type of another person given two images

Set-I

Set-II

Workflow

Requirements

• OpenCV
• scipy
• numba
• numpy
• mediapie

Usage

pip install -r requirements.txt

python seam_carving.py (-resize) -im1 <IM_PATH> -im2 <IM_PATH> 
                         -out <OUTPUT_IM_NAME> [-dx <DX>] 

Description of parameters:

• -im1: The path to the image to change.
• -im2: The path of the target image.
• -out: The name for the output image.
• '-dx- : Number of pixels to remove. Positive for adding pixels, negative for subtracting pixels. Optional. If not present number of pixels gets calculated.

Algorithm used:

1. Mediapie library used for key point detection
2. Left and right shoulders detected, distances between shoulder measured. Say distance between lt and rt shoulder for person 1 is 15 while for person 2 it is 20. So if 5 pixels are removed from the middle of the person then we get person 2 looking like person 1
3. The difference in above distance, the number of pixels removed as the seam through an algorithm called seam carving algorithm

Key-point Detection

To detect essential features of the images to determine by what amount to reduce or increase the width of the given image

Seam-carving Algorithm

An effective method to resize images by removing pixels without cropping images or resulting in loss of information

Acknowledgements

Many parts of the code are adapted/optimized versions of functionality from other implementations:

• https://github.com/axu2/improved-seam-carving
• https://github.com/vivianhylee/seam-carving
• https://karthikkaranth.me/blog/implementing-seam-carving-with-python/
• Seam carving for content-aware image resizing(2007) link
• Mediapie for key point detection
• A fast Python implementation of Seam Carving for Content-Aware Image Resizing (2007), including the improved energy algorithm described in Improved Seam Carving for Video Retargeting (2008).