Query by Approximate Shapes Image Retrieval improved object sketch extraction algorithm

In this paper a new Content Based Image Retrieval based on a sketch method was proposed. The main idea of the algorithm is based on decomposing an object into predefined set of shapes (primitives): line segments, polylines, polygons, arches, polyarches and arc-sided polygons. All primitives are stored as a graph in order to store the mutual relations between them. Graphs are stored in a tree-based structure which allows fast querying. As an improvement to the algorithm, a conversion to the HSL color space was proposed in order to detect primitives more accurately. Moreover, computing all line slopes in relation to the object oriented bounding box was also proposed. Additionally, in order to better detect objects present in images, the usage of Edge Boxes algorithm was proposed.