Multi-Source Heterogeneous Iris Recognition Using Stacked Convolutional Deep Belief Networks-Deep Belief Network Model

Deep Learning for Iris Recognition: A Review

“…The image segmentation in the iris identification task aims to segment the iris region in people's eyes accurately. Traditionally, classical threshold-based segmentation is used to perform this task, which aims to seek a suitable threshold to binaries the image based on the lower grey value of the pupil [17]. However, a single threshold value makes the segmentation method lack generalization ability.…”

“…However, a single threshold value makes the segmentation method lack generalization ability. In order to make the iris images obtained in different environments well segmented, the Hough transform model has become a more commonly used method in iris segmentation in recent years [17][18][19][20][21][22].…”

“…These studies applied the Hough transform model directly to Region of Interest (RoI) to localize the iris efficiently. Differently, [17] employs the binarization method to locate the inner boundary points and the Hough transform to discover the inner boundary circles for iris segmentation.…”

“…7 illustrates the effect of this removal operation: The left is the original iris image, and the right is the processed image by removal of specular highlights via linear interpolation. Some studies also focus on image enhancement methods [17,19,20,30]. [17,19] use the histogram equalization method.…”

“…Some studies also focus on image enhancement methods [17,19,20,30]. [17,19] use the histogram equalization method. The cumulative distribution function enhances the contrast of the image without changing the original brightness space of the image.…”