Eğricik Dönüşümü Kullanarak El Yazısı İmza İmgelerinin Erişiminde Mesafe Yöntemlerinin Analizi

This paper presents the effect of distance metrics on curvelet transform based handwritten signature image retrieval. A total of 240 signature images obtained from 20 different people were used as the image database. The mean and standart deviation values of each subband matrices formed by applying cuevelet transformation to signature images were used as features in retrieval process. Then, the comparison of these features was made by using different distance calculation methods used in other studies on curvelet transform. Compared distance calculation methods are euclidean distance, manhattan distance,

chebyshev distance, canberra distance and jeffrey distance. Each image in the database was used as a query image and other signatures made by the same person were tried to be determined. The comparison of performance has been achieved by averaging all results over precision and recall curves. As a result, Jeffrey divergence method was determined as the most successful method among the compared methods