Most image tamper detection and localization schemes in the present day rely on huge volumes of training data to achieve perfect performance. To the best of our knowledge, the state-of-the-art schemes rely on thousands of training samples (ranging from 1K to over 84K) to localize forged image regions. In this work, we aim to come around the problem of reliance on huge volumes of training images in order to efficiently locate a tampered image region; in fact, we succeed to achieve zero training data dependency for image tamper localization. In this paper, we have shown our work on a specific class of image forgery, viz. image splicing attack. To state more specifically, in this paper, we propose a set of optimal image features, which are subsequently fed to a hierarchical agglomerative clustering module, thereby detecting and localizing spliced region(s) within an image. Our experiments prove that the proposed method achieves close to 90% accuracy while completely bypassing any training data requirements and solely relying on the unsupervised clustering concept.