Signature is one of the most important and widely accepted biometric modality. It is the most common biometric used in documents like financial transactions, legal documents, contracts, etc. Over the years, many signature verification methods have been proposed; however, it is a common notion in most of these methods that signature is available separately for verification purposes. In real world scenarios, signatures are not always available separately particularly in forensics. In documents, signatures usually overlap with other parts of the document, like printed text, lines and graphics, where it becomes practically impossible to detect and localize the signature pixels. In this paper, we present a robust and very effective method for signature segmentation from documents using hyperspectral imaging. A comparative analysis of state of the art key-point detection based method and proposed hyperspectral unmixing method are provided. The preliminary study shows that spectral unmixing offers great potential for automatic signature extraction from document images.