Compression of Large-Scale Image Dataset using Principal Component Analysis and K-means Clustering

Abstract

Digital images, being on the verge of its utmost

popularity encompasses plenty of applications and as such are

generated at an unprecedented rate. These digital form of data are

often found with redundant information. Applications that

require a bulk amount of images to be processed, turn out to be

high regarding computational complexity. Needless to say, it leads

to inefficient storage utilization. In this paper, a hybrid approach

is applied to compress a large-scale image data-set by combining

two popular algorithms: Principal Component Analysis (PCA)

and K-means. This paper works with a view to diminishing the

redundant information by implementing dimensionality reduction

followed by color quantization. The PCA is used to project the data

onto a lower dimensional space with retaining as maximum

variance as possible. The K-means algorithm is used to restrict the

distinct number of colors to represent an image by means of

clustering the data together. The results obtained from the

proposed method is compared with the results obtained from

implementing PCA and K-means clustering algorithms

independently, where the proposed method provides with a better

compression ratio.