Local Mean Directional Intensity Pattern: An Efficient Descriptor for Hand Gesture Recognition Using SVM Classification

Hand Gesture Recognition (HGR) promotes the most efficient and natural communication between users and machines. However, the performance of the HGR system is deteriorating due to challenging conditions such as illumination variation, complex backgrounds, inter-class similarities. In this paper, a handcrafted feature descriptor named as Local Mean Directional Intensity Patterns (LMDIP) is proposed for effective hand gesture recognition in the aforesaid challenging environment. The proposed LMDIP descriptor aims to maintain the most relevant directional information. It assists in the detection of grey level variations that move in different directions. LMDIP explores differential excitation by calculating the mean intensity of pixels across each angular neighboring pixel. The most significant effect of employing the mutually beneficial connections between adjacent neighbors in different angular directions is that it no longer relies solely on the sign of the intensity differences between the center pixel and one of its neighbors, but also on the sign of the gap across its adjoined neighbors. The performance of the proposed LMDIP is evaluated in terms of accuracy and F1-score on five benchmark datasets: Massey University Dataset (MUGD), ASL Digit Datasets, ASL Static Dataset, NUS Dataset, and OUhand Datasets. The result of the experimental show that the proposed LMDIP achieves an accuracy of 92% (MUGD Set1), 98% (MUGD Set2), 75%(MUGD Set3), 61%(MUGD Set4), 71%(MUGD Set5),99%(ASL Digit), 51%(ASL Static), 83%(NUS-I Dataset), 39% (OUhands Dataset) and respectively which is better compared to the state-of-art methods.