Query by Approximate Shapes Image Retrieval with improved object sketch extraction algorithm

Abstract

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.

QoS-oriented capacity planning for edge computing

“…They designed a mapping system called CrowdMap, which involves a 2-step image matching process. In the first step, the extracted frames are quickly matched using color-indexing histograms [130], shape matching [42] and wavelet decomposition [24]. Since the techniques used in the first step are sensitive to illumination changes and camera parameters, in the second step, the matched results are refined using SURF features [27] for precise matching.…”