Image classification based on focus

The performance of most image classification algorithms deteriorates in the presence of out-of-focus blur. Thus, it is essential to either correct the focus of the input images or leave them out of the training set. There exist many focus metrics for auto-focusing, but they generally give a relative focus value. Our technique combines some of the best performing focus metrics to obtain a new focus measure using which we can separate in-focus images from out-of-focus ones. We also compare our technique with the existing ones and show that it performs better. The classifier was tested on a dataset of ovarian images obtained using confocal microendoscopy.