chandu-97/eigenfaces

Face recognition using eigenfaces

We present an approach to the detection and identification of human faces and describe a working, near realtime face recognition system which recognizes the person by comparing characteristics of the face to those of known individuals.

Face detection

Viola Jones detector which is used for face detection performs much better and can detect faces in real time. It can detect faces irrespective of their scale and position. A result is shown here by using the image of George Bush. This is not cherry picked but just a random image on the Internet.

Methods used:

Independent component analysis

The top 12 ica components are shown here below.The Image does not look as much intuitive as PCA eigen faces.But the structure and relative position of nose, eye’s, eye brow’s, lips etc; is still maintained

Non negative matrix factorization

The highest accuracy for NMF can achieved with projecting onto 219 components and the accuracy is 82.9%.This is comparable to PCA and ICA but lesser than both of them.

Local Linear embedding

The Highest accuracy for LLE is with number of components as 26 and number of neighbour’s as 27.The accuracy is 33.85%.It is not better than PCA, ICA and NMF.

t-Distributed Stochastic Neighbor Embedding

The highest accuracy for TSNE is 45% is almost constant with the increase in dimension’s to which it is being projected onto.The class output is same disregard of the input.

Report

Report link