PCA-SVM-Features-based-Face-Images-Classification

Objectives:

Face recognition on The Database of Faces at a glance.
Data pre-processing.
Dimension reduction using PCA.
Support Vector Machines (For classification)

Project work flow:

Data-Set�(The Database of Faces at a Glance):

The data-set consists of 400 pictures of 40 people.
Each picture has a dimension of 112X92 pixels.
Data set comprises of pgm format.

Front faced pictures of a single subject was captured in different perspective & can be seen by plotting few examples of the given data-set.

Data pre-processing:

Each image is flattened into row vector of size 10304.
The size of whole data set becomes 400X10304.
Then converted into a numpy ndarray.

Dimension reduction using PCA:

Redundant features are removed using PCA.
Eigen vectors with more variance are kept while with low variance are discarded. 
Output matrix from PCA will look noting like original image.

Support Vector Machine�(SVM):

First of all for classification SVM is applied directly without using SVM.
To get maximum margin, Linear classifier was used.
100% accuracy was achieved.

Score without using PCA:

When original numbers are used that is dimension reduction isn’t performed we have maximum score.

SVM after using PCA:

>PCA was implemented for K=1,2……10
>For K=1 , error was 94%.
>For K=10, error was 1%.
>Significant improvements made when dimensions were increased.

Results with PCA:

As we are increasing number of dimensions error is decreasing and score is increasing to one.