Biometrics refers to the automatic identification of a person based on his/her physiological or behavioral characteristics. A biometric system is essentially a pattern recognition system which recognizes a user by determining the authenticity of a specific characteristic possessed by the user. Extracting unique features from the human body is an important task. Curvilinear structure is one of the most popular features used in biometric systems. However, even though current techniques exist to extract line features, none retain its wide information well, which is necessary in biometrics. In this thesis we propose an approach to solve this problem. After analyzing the cross-sections of given lines, we notice they are Gaussian shaped. Hence, a Gaussian filter to match them is suitable to be used. Applying a single scale filter generates a lot of noise and/or loses details. To overcome this deficiency we develop a multi-scale approach, i.e., three scales according to the cross-section widths (largest, smallest and average) as well as eight directions (horizontal, vertical and diagonal). A response is calculated by convoluting the original image with the filter. Two responses using different scales and directions form a production which exhibits less noise than a single scale filter and also preserves wide information. Two biometric applications are applied to illustrate the effectiveness of our approach, one is for personal authentication using palm veins and another for recognizing Diabetic Retinopathy (DR) in retinal blood vessels.