Informatics Publishing Limited

Nahed Tawfik ¹, Mahmoud Fakhr ², Moawad I. Dessouky ³

Affiliations
1 Computers and Systems Department, The Electronics Research Institute, EG
2 Electronics and Communications Department, Cairo University, IN
3 Department of Electronics and Electrical Communications, Menoufia University, IN

Abstract

This paper presents a proposed method for corneal pattern classification using a Cepstral approach and SVMs. This approach based on the transformation of the corneal image to 1D signal, the feature extraction process and finally the classification process. MFCCs are one of the best feature extraction techniques used in 1D signal. This approach composed of two phases: a training phase and testing phase. In the first phase, a database of the corneal patterns is applied to obtain features from each corneal image, and then these features are used to train Support Vector Machines. In the second phase, features are extracted with the same steps in training phase from a set of new corneal images and finally a feature matching process is carried out. In this work, 1D signal used with time domain or in different discrete transform domains. The experimental results indicate that this technique achieves high classification rate up to about 100%.

Keywords

Corneal Images, MFCCs, Support Vector Machines (SVMs), DCT, DST, and DWT.

Full Text

     

Nahed Tawfik ¹, Mahmoud Fakhr ¹, Al-Nuaimy Waleed ², Moawad I. Dessouky ³, Fathi E. Abd EI-Samie ⁴

Affiliations
1 Computers and Systems Department, The Electronic Research Institute, EG
2 Department of Electrical and Electronic Engineering, University of Liverpool, GB
3 Department of Electronics and Electrical Communications, Menoufia University, EG
4 Menoufia University, Menouf, EG

Abstract

In this paper, a new technique for feature extraction from corneal images is presented which can be applied for corneal pattern recognition. Most of the previous methods are based on segmentation of the corneal images which are restricted to certain planes. In this paper, a proposed method is applied on corneal images which have two main phases. Firstly, the 2-D images are lexicographically ordered to 1-D signals, and then the Mel Frequency Cepstral Coefficients (MFCCs) and polynomial coefficients are extracted from these 1-D signals or from their transforms. Secondly, the SVM is used to match the extracted features in the testing phase to those of the training phase. Experimental results show that the recognition rate for features extracted from Discrete Sine Transform DST and Discrete Cosine Transform (DCT) achieve better performance compared to other cases. The method in this paper is limited to feature extraction for pattern recognition and the automatic diagnosis case is left for future work.

Keywords

Corneal Images, Pattern Recognition, Mel Frequency Cepstral Coefficients (MFCCs), Polynomial Coefficients, Support Vector Machine (SVM), Discrete Cosine Transforms (DCT), Discrete Sine Transforms (DST), Discrete Wavelet Transforms (DWT).

Full Text