Artificial Neural Network Based Spermatozoa Classification Using First Order Statistics and GLCM Features

Spermatozoa morphology is one of the main characteristics used for evaluating semen fertilizing capacity. This paper aims at classifying the morphological assessment of each spermatozoon images obtained from WHO laboratory manual either as normal or abnormal. Images were cropped and resized to 86 x 100 pixels. The resized images are segmented using a threshold based method. The texture of the segmented image segment is evaluated based on the Gray level Co-occurrence matrix (GLCM) and first order statistics (FOS) features are extracted. GLCM for the segmented gray scale image were calculated in 4 angles (0, 45, 90 and 135) at an offset of. Totally 15 GLCM features and 4 FOS features are extracted. The extracted features are then used to train and test the artificial neural network constructed using Feed Forward Neural Network, Radial Basis Neural Network and Elman Back Propagation Neural Network. Experimental results are presented on a dataset of 91 images consisting of 71 abnormal images and 20 normal images. The classification accuracy of 75% is achieved when feed forward neural network is trained with GLCM features, 58% when Recurrent network is trained with FOS features and 75% when Radial basis neural network is trained with the combined features (GLCM+FOS).