Detection of Cholesterol Using Eye Picture Based Local Binary Pattern (LBP) and Support Vector Machine (SVM)


Cholesterol is a fat in human blood that is needed for the formation of several hormones and new cell walls. Normal human cholesterol levels are in the range of 200 mg / dL or less, but if above 240 mg / dL will be at high risk of various dangerous diseases such as stroke and coronary heart disease. If cholesterol levels are not detected early, the risk of stroke and coronary heart disease is very large, considering that coronary heart disease is one of the many diseases that cause death. This study designed a system that can be used for early detection of cholesterol levels with a short time through the eye image.

After the data acquisition process then the eye image data is carried out by a preprocessing process which consists of the resize process, ROI circle crop, and conversion of RGB eye images to grayscale. In this study the Local Binnary Pattern (LBP) method is used as a feature extraction method by recognizing images based on patterns to look for certain traits in an image that will be stored as traits of training images. And using the Support Vector Machine (SVM) classification method to find classes for each test data including the normal class or cholesterol class.


In this study a system has been designed for the classification of cholesterol levels through eye images using the LBP method as feature extraction and the SVM method as a classification. The system can classify cholesterol levels based on 2 conditions namely normal conditions and high cholesterol conditions. Based on testing that has been done the parameters of ROI 60 and 62 have the highest accuracy with an accuracy of 90% with the fastest computing time with an average of 0.376 seconds. The parameter which also influences is the radius where the greater the radius value the smaller the accuracy, the radius value 1 and 2 have the highest accuracy of 90%. Without using statistical features Linear kernels have the highest accuracy 85-90% better than Gaussian and Polynomial kernels. Image resizing also affects the accuracy of the 512 X 512 pixel image resizing system which has the highest accuracy. Statistical characteristics have an influence on the accuracy value where the combination of gaussian kernel statistical characteristics with as many as 30 image test data has an accuracy of 93.33%.

e-Proceeding of Engineering : Vol.6, No.2 Agustus 2019 | Page 3814

Telkom University – Bandung , Indonesiaia

Dimas Prihadi Waluya1,

Efri Suhartono,

Irma Safitri