The Research Of Classification Method Of Iris Image Quality Based On BP Neural Network

Iris image quality has attracted more and more attention. A poor quality image canseverely affect the iris recognition matching efficiency. Compared to the currently widelyused identification methods, iris recognition has the advantages of uniqueness, reliabilityand non-invasive, thus it has the incomparable advantage over general identificationmethods. Researchers in this field have also achieved some results. Iris imageclassification can solve well the problem between the recognition speed and recognitionaccuracy. The classification of iris image quality is to prevent the bad quality iris imagecoming into the subsequent recognition system, which affects the efficiency of the irissystem identification. The theory of iris image quality evaluation has been developed inrecent ten years, because the image quality is affected by various factors, thecharacteristics of it are variously not ideal, there are not an unified judgment criteria foriris image quality. The thesis proposed a new iris image quality classification method,basing on the understanding of the existing research methods.In the process of iris image acquisition, the iris image quality is not ideal, because ofthe influence of the insufficient light conditions, glasses reflections and eyelid, etc. Thisthesis puts forward a new classification method based on the back propagation (BP)neural network. It applies wavelet transform to the iris image feature extraction, and theninput the extracted normalized iris image data to the BP neural network. After the BPneural network is trained by the data, it is able to distinguish that weather the iris image isaffected by one of the three factors or not. The experimental results show that the methodhas high iris recognition performance and low error rate. This algorithm combinedwavelet transform and BP neural network, at the same time, this thesis also didexperiments by other algorithms. The proposed method in this thesis not only enriches the study of the iris recognitiontechnology, but also can distinguish the bad quality images with normal imagesaccurately. The abnormal images in this thesis include three kind of situation: the irisimage which in poor light conditions, the iris images which in glasses reflection and theiris images which in eyelids and eyelashes. Simulation results show that the algorithm hasgood accuracy in the assessment of iris image quality, and it has a certain theoretical andpractical significance. Actually, there are still some defects in the algorithm. Iris imagequality assessment is based on human subjective evaluation, so it is only applicable to theiris database of this study, which needs further research and improvement.In this thesis, the innovation points are as follows:1) The wavelet coefficients are regarded as the quality evaluation of iris imagecharacteristic. Wavelet transform is commonly used in image analysis. This thesis usesthe wavelet coefficient after wavelet decomposition to describe the texture feature of irisimage, which is taken as the BP neural input, and the algorithm is simple.2) The choice of the training samples. The selection of training samples for BP neuralnetwork is based on human subjective judgement. Training samples are the sampleswhose quality is known. The quality of the iris image is decided by people’s subjectivejudgement, and we apply the judgement criteria to the training samples.