Image Quantization using Kmeans and Image compression using PCA

Pre-requsite:

Make sure you have anaconda, scikit-learn and scikit-image

1. Color Quantization

Used kmeans clustering to create centroid from sample of 1000 points in image array / raster. Then predicted center for each point in the raster and ploted them which resulted in quantized image of the original image

How to Run:

1. run python quantize.py <imagefilePath> <valueOfK>

   example python quantize.py images/image3.jpg 32

Output:

• Output images are stored in folder quantizedImages as filename-<kvalue>k.jpg

Report:

Image	Original File Size	Value of k used	Image Quality	Image Size
Image1	355.4K	2	Just two colors visible sort of black and light blue	302.4K
Image1	355.4K	8	Some color details missing in the sky	334K
Image1	355.4K	16	Almost similar	334.4K
Image2	536.6K	8	Object edges are blur	292.1K
Image2	536.6K	16	Blur dark colors	296.9K
Image2	536.6K	64	Almost similar	285.9K
Image3	598.4k	16	Blur	402.4k
Image3	598.4k	32	Less Blur	411.6K
Image3	598.4k	64	Almost similar	414.3K

References

• https://scikit-learn.org/stable/auto_examples/cluster/plot_color_quantization.html
• https://en.wikipedia.org/wiki/Color_quantization
• https://lmcaraig.com/color-quantization-using-k-means/
• https://appliedmachinelearning.blog/2017/03/08/image-compression-using-k-means-clustering/

2. PCA for Images

Ran PCA with different number of components. Transformed input array of every color channel i.e. for Red, Green and Blue into pca components. Then inverse transformed pca components to array. Plotted this arrays of all channel to for compressed images.

How to Run:

1. cd into path/to/partIII

2. run python compress.py <imagefilePath> <valueOfPrincipalComponents>

   example python compress.py images/image3.jpg 100

Output:

• Output images are stored in folder compressedImages as filename-<pcvalue>pca.jpg

Report:

Image	Original File Size	Value of Principal Components	Image Quality	Image Size
Image1	355.4K	100	Multi-colored noise	480.1K
Image1	355.4K	150	Multi-colored noise	514.1K
Image1	355.4K	200	Multi-colored noise	532.1K
Image2	536.6K	150	Noise in white color	292.1K
Image2	536.6K	200	Less Noise in white color	296.9K
Image2	536.6K	500	Almost similar with un-noticable noise in white color	285.9K
Image3	598.4k	800	Hazy with noise in white color	402.4k
Image3	598.4k	1000	Hazy	411.6K
Image3	598.4k	1200	Almost similar to original	414.3K

References

• http://people.ciirc.cvut.cz/~hlavac/TeachPresEn/11ImageProc/15PCA.pdf
• http://www.scielo.br/scielo.php?script=sci_arttext&pid=S1679-45082012000200004
• http://dilloncamp.com/projects/pca.html