Trace transform based identifier for speech based image retrieval on mobile phones

Present day mobile phones have evolved as multimedia devices, where users can capture and store photos, videos on their mobile phones. As the amount of digital multimedia content expands, it becomes increasingly difficult to find specific images in the device, giving rise to problems of organization, storing and retrieval of images. To improve human access to a large unstructured data in their personal collections on mobile phones, there is a need for effective and precise retrieval algorithms for the user to search browse and interact with these collections in real time. Retrieval algorithms are highly complex and this characteristic becomes more intense on mobile platform due to restrictions in architecture and computing power. In this paper we propose a speech based image retrieval algorithm for personal collections optimized for porting on to a mobile phones. We have treated the speech spectrogram as an image and applied trace transformation to obtain an unique and robust identifier string that acts as a fingerprint for image retrieval systems. Trace transform is popular in image processing algorithms because it is robust to affine transforms for feature extraction. The proposed algorithm exhibits optimization in memory and retrieval time costs.