Research On Face And Iris Matching Layer Fuzzy Fusion System

There are many types of biometric identification systems, amongst them fingerprint identification, face recognition, iris recognition, voice recognition, hand shape recognition, palmprint recognition, gait recognition, and signature recognition are some of the more common biometric identification systems. Identification systems that use only one kind of biometric features are called mono-modal biometric systems. Though each biometric feature has their own strengths and weaknesses, it is often found that performances of mono-modal systems are rather limited and are more vulnerable to spoof attacks. To overcome the shortcomings of mono-modal biometric systems the idea of multimodal biometric systems emerged. Multimodal biometric recognition could mean a system that conducts recognition using a single biometric feature but with multiple integration (for example, face images collected at different angles circumstances), or it can be to integrate a variety of different biological characteristics (such as face, iris, or fingerprint).Face and iris recognition systems have significantly improved over the last 20 years. In fact with all the research and funding that this field of study has received within the last decade, it is understandable how it is possible for both face and iris recognition systems to reach recognition rates as high up as 95% in actual field tests. But as mentioned before, performances of mono-modal systems are rather limited and are more vulnerable to spoof attacks due to limited feature spaces. Therefore through careful observation and extensive research on both face and iris recognition systems, one would realize that both face and iris recognition systems are highly dependent on the quality of the image itself. Any form of noise that appears in the input image can drastically decrease both face and iris recognition systems, especially iris recognition systems, which requires iris images of exceptional quality in order to maintain a high recognition rate. In cases where low quality images are obtained current face and iris recognition systems will display a "Failure to Enroll" message and inquire the user to re-enter their biological features, though this is a way to maintain the recognition rates high, it is an inconvenience for users to enter their biometric feature over and over again until a perfect image is obtained. The aim of this abstract is to create a multimodal identification system for face and iris, but before going into further details on the system itself; one must ask what our goals for the multimodal identification system are. First of all the multi-modal identification system must have a higher recognition rate than their original mono-modal identification system. In addition, in a worst-case scenario situation, where low quality images are used as input images, the multi-modal identification system we propose must obtain a recognition rate higher than the mono-modal identification systems.First, the paper starts off by introducing the methods we used for both facial feature extraction and iris feature extraction, then give an overview on Zadeh's fuzzy sets and introduced the concept of fuzzy systems. Finally, in-depth research on V.Conti's fingerprint recognition system and its fuzzy fusion system is conducted. Here the paper analyzes V. Conti's fuzzy fusion system and show how his theory of Goodness Index variable (a variable designed to determine the image quality) can be used to conduct fuzzy fusion of two biometric features.As V. Conti's Goodness Index variable only applies to determining the image quality of fingerprint images, modifications to the original Goodness Index variable had to be made in order for it to work on face and iris images. Through extensive research on face and iris test results, the paper proposes altogether 4 factors (2 factors for facial images and 2 for iris images) that affect the quality of face and iris images. After determining the factors, equivalent fuzzy inputs for these factors have to be designed, through an understanding of Zadeh fuzzy sets and Zadeh's fuzzy system, this article have devised prototype fuzzy inputs for each of the four factors. Tests and experiments are then conducted to test out these prototype fuzzy inputs to determine its fuzzy parameters.After acquiring 4 fuzzy inputs, there still lies the problem of fusing them together into one single variable or aka. Goodness Index variable. Since the fuzzy inputs consists of both Interval Type-2 fuzzy sets and Type-1 fuzzy sets, a Type-Reducer has to be included in order to fuse both Type-1 and Type-2 fuzzy sets together. In this instance, Wu-Mendel's Type-Reduction and defuzzifier algorithm are used to fuse all the fuzzy inputs together. Then through establishing fuzzy rules that combines the Goodness Index and matching scores for both iris and face, a fuzzy recognition system on a matching score level is created. The fuzzy system uses the knowledge base built with the fuzzy rules we design for a face and iris recognition system. Experiments and tests are then conducted on the face and iris fuzzy fusion recognition system and results show our multi-modal recognition system has significant improvements over its mono-modal recognition systems.