Research On Planar Footprint Shape Analysis And Footprint-based Identity Verification

Biometrics exploits physical features or behavioral features to identify a person. It has been becoming one of the research hotspots in information domain. Mark on the ground left by human soles under body weight is called footprint, and it is a type of important trace information belongs to biometrics. Abundant expert experience on footprint verification has been accumulated especially through long time of forensic application, but the research on footprint verification has far from sufficiency in being considered as a information technology compared with other human biometrics. There is still a mass of fundamental and application problems to deal with.Under the condition of standing or walking, a human body forces its sole surface onto a flat rigid ground and makes a planar impression. In an image of planar barefoot impression, there are areas and edges that can reflect both physical and behavioral features in a body, and can persist uniqueness and stability in the same circumstances, they are just the main study content in identity verification based on planar barefoot impressions. Using experts' experience as a reference and in view of the inherent characteristics of human footprint, this thesis gives a research on the key technologies and algorithms of footprint image shape analysis, including footprint image collection, image processing, footprint feature extraction, and addresses the scientificity of identity verification based on planar barefoot images. The main work and contributions are summarized as follows:1. Research on segmention of planar barefoot impression images. As the edge of weight-bearing areas in the footprint image is always located in the vicinity of strong edge, constrained by a certain shape mode and smooth locally, it proposes a boundary finding algorithm based on DFD(Discrete Fourier Descriptors) parametrically deformable model. This algorithm constructs a DFD parametrically deformable model to describe a boundary, and defines a optimize function of total income of gray fuzzy consistency in the area. Then the boundary of weight-bearing areas are reconstructed by the DFD parametrically model by applying genetic algorithm on the optimize function and getting the optimized parameter set. Experimental results show that this method can extract the weight-bearing area boundary in an effectively way by keeping the feature of the certain shape mode as well as describing the detail in the boundary curve.2. Research on discrete curvature estimation of 2D curves. For the purpose of describe the curve properties of boundary in the barefoot impression images in an effective way, it proposes a method of discrete curvature estimation based on adaptive fuzzy segments for 8-connected curve in 2D space. This algorithm considers the infection made by discrete noise on 2D curves and chooses a proper "order" of fuzzy segment according to the local curve coarse degree of curve adaptively, then curvature on each point is estimated by its tangents approximated by the longest fuzzy segments grown from points on the curve with the given order. Experiment results show that our method improves the performance of curvature estimation in keypoints' detection as well as having a better consistency with the feature in continuous space.3. Study on the configuration of footprint and geometry features' extraction. In order to locate the dominate points on the boundary of footprint, such as the top point of the heel, the toes, the ending points of metatarsus, it provides methods of setting up reference axis for footprint images, which can also make footprint's shape immune to rolling. The whole boundary of planar barefoot is divided into curve segments as the edge of metatarsus, outside and inside arch, etc. A set of geometric feature measurements are extracted to describe the structure and geometric properties of barefoot, and the relationship between each pair of them is discovered as well as the stableness and uniqueness for each feature is compared.4. Study on the feature extraction for footprint based on areas and boundaries. The main clue to verify one footprint from the other is the shape difference between them. After applying a analysis on the shape characteristics of the areas and boundaries in footprints, through the methods of defining a set of feature measurements computed from area, such as the footprint support degree(SD), the footprint by-press degree(BD) and the arch press degree(APD), adapting the polynomial models to describe the segmented curves, and etc., it set up a feature vector to reflect the shape properties for each barefoot impression. the stableness and uniqueness for each feature in this vector is also compared.5. Investigation on identity verification methods based on planar barefoot impressions. Construct a identity verification system base on planar barefoot impressions. In this system, we establish a footprint verification probabilities model, make a feature selection using sequence backward method according to the disperse separability criterion, and then adapt a decision rule for footprint verification considering its specific request in application. The result of the experiments on a set of footprint samples with 398 different objects show that the eigenvector extracted to describe the shape properties of footprint is effective and the model of footprint has a good performance by achieving a 93.91% correct rate.