1 (d) Given two lines in a palmprint image, please explain how to get their Euclidean distance (P10: 21). If we mark each line as orientation features, how to understand their matching score: S(C1, C2) in P10:33?
2. In line feature based offline palmprint recognition, four different directional templates are defined. Please try to explain why they can detect the corresponding directional line segments (See P10:19). Can you design another set of four different directional templates to determine line segments?
A line segment 10 90 10 90 10 90 10 90 10 Its cross-section 10 90 10 Our line detector Its cross-section 121 The inner product of two vectors increases with their similarity !
Line detector Zero sum Similar cross-section to a line segment Larger than a line segment The better similarity, the better detection How many directions we need? The more directions, the more accuracy, the less efficiency A tradeoff is made according to experience
3. In P10:28-29, a segmentation approach, Tangent Points of the Finger Hole, is introduced. The line passing though the two points, (x1, y1) and (x2, y2), satisfies the inequality, for all i and j. Please understand this method and check the determinant rule.
Tut 7, 2. There are four steps in the Daugman’s approach (P8: 32-36). The third step generates IrisCode with 512 bytes. If 2 bits represent a feature, please compute the total number of features. (512*(8/2)=2,048)
5. In texture features matching stage, hamming distance is used to measure the similarity of two palmprints, as shown in P10:37. What conclusion can you get from the normalized distance D 0 ? Hint: Consider the normalized distance with the experimental results in P10:38. (The smaller in the normalized distance, the closer in two palmprints)