This paper presents a new approach to address the problem of offline handwritten signature verification. In contrast to many existing systems, we are interested in making soft decision rather than a purely binary classification for the signatures under verification. To accomplish this goal, we incorporate both types of features: finer intensity-based features and global geometry-based features. Particularly, the finer features are computed for every sample point of a signature using histogram of intensities, and the geometry-based features are extracted using an adaptation of the shape context descriptor. One of the advantages of our approach is that the extracted features are very robust to noise, rotation and scaling change without heavily relying on any complicated pre-processing steps. The extracted features are used to compute the similarity score, followed by a score calibration process to estimate the corresponding confidence score (i.e., using log-likelihood-ratio). To validate...