Robust Feature Descriptor Employing Square Triangle Tessellation for Shape Retrieval

Recent studies on shape retrieval stress for the realization of highly efficient feature descriptors realized with reduced complexity. Accordingly, a simple tessellation operation that geometrically explores the spatial data for realizing efficient and precise shape descriptor is dealt in this paper. The descriptor labelled as Squared-Triangle Tessellation Descriptor (STTD), enforces strict geometrical congruency to facilitate effective feature extraction and representation. STTD dually tessellates the image into square tiles and later decomposes them into triangles. Upon triangle formulation the respective features are capitulated using simple geometrical means which is then transformed into a shape histogram. Then, an auto encoder operates on the constructed feature database and classifies the diverse shapes based on the intra and inter-class relationship that exist amongst the different features. Exhaustive investigations on publicly available dataset namely MPEG7 Part B, Tari-1000 and Kimia’s 99 reveal consistent accuracy of 99% offered by STTD across these datasets when compared with its competitors. As majority of the STTD formulation deals with integer arithmetic therefore simple multipliers with less area and power is suffice for its VLSI implementation, thereby, amicable for real-time applications.