Iris recognition using fully convolutional networks
A method of accelerated iris recognition includes acquiring an image comprising at least an iris and a pupil, segmenting the iris and the pupil using a fully convolutional network (FCN) model, normalizing the segmented iris, encoding the normalized iris, the normalizing and encoding using a rubber sheet model and 1-D log Gabor filter, and masking the encoded iris.
1. A method of accelerated iris recognition comprising:
acquiring an image comprising at least an iris and a pupil;
segmenting the iris and the pupil using a fully convolutional network (FCN) model and a circle fitting algorithm;
normalizing the segmented iris;
encoding the normalized iris, the normalizing and encoding using a rubber sheet model and 1-D log Gabor filter; and
masking the encoded iris.
2. A method of accelerated iris recognition comprising:
exploring fully convolutional network (FCN) architectures for iris segmentation;
evaluating a performance versus a complexity trade-off for each FCN architecture by executing a full end-to-end iris recognition pipeline;
performing FCN selection based on its complexity and its end-to-end iris recognition performance such as equal error rate and receiver operating characteristics;
evaluating FCN complexity by measuring parameter counts and execution latency on an accelerator running on an embedded field programmable gate array (FPGA) platform; and
executing a full pipeline implementation on an embedded field programmable gate array (FPGA) platform.
3. The method of accelerated iris recognition of claim 2 wherein iris segmentation comprises:
an accurate circle fitting algorithm that computes center points and radii of the pupil and limbic boundaries from a segmented mask.
4. The method of iris accelerated recognition of claim 3 wherein the iris recognition processing pipeline comprises a hardware accelerator design for FCN models that uses a combination of CPU vectorization and hardware acceleration.