Abstract
The emerging field of graph signal processing has brought new scope in understanding the spectral properties of arbitrary structures. This paper proposes a novel graph spectral domain feature representation scheme for recognising in-air drawn numbers. It provides the solution by forming the hand's path as a graph and extracting its features based on the spectral domain representation by computing the graph spectral transform. A novel graph generation model is proposed to form the topology of the shapes of numbers. The experiments show that the proposed features are flip and rotation-invariant which makes insensitive to changes in the rotation angle of the drawn numbers. The proposed solution achieves a high level of accuracy of nearly 98% for in-air hand drawn number recognition.
Highlights
Irregular data and complex structures are rapidly increasing number of data due to the vast improvement in the technology
This paper suggests utilizing two types of feature as follows: 1) First moment components (Mi) there is a similarity in the second eigenvector vector between samples, a small difference details is shown at the beginning or at the end of the Fiedler vector such as number two and five
Graph is a new mechanism to deal with unstructured data that provides an optimal description of the geometric structure
Summary
Irregular data and complex structures are rapidly increasing number of data due to the vast improvement in the technology. The proposed method falls into two areas, which are complex shapes matching field and in-air hand gesture recognition. The massive time requirement for implementation makes these approaches unsuitable for real-time applications To address these limitations, this paper proposes feature-to-feature assignments, which is based on the geometric structure properties of the objects. Our method is totally different than the existing works in the fields of graph matching and hand gestures recognition. The use of graph in the hand gesture recognition field is very limited and restricted by identifying hand poses only, which makes this paper the first to apply the spectral graph feature for in-air hand drawn numbers recognition. The key point of these works is how to allocate the graph nodes over the hand based on features.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have