Indian Cultural event detection using GBVS boosted deep convolutional neural networks

Abstract

Cultural heritage plays a significant role in shaping the identity of a region but recognizing cultural events poses a notable challenge in the field of computer vision. In this paper, a novel framework is proposed that blends convolutional neural networks (CNNs) and spatial feature learning for identifying Indian cultural events from image samples. The proposed method first starts by extracting spatial characteristics that capture important information about cultural events using graph-based visual saliency. These extracted images were then fed to a DenseNet201 model for event recognition. This model had an SVM classifier as the last model layer instead of the Softmax layer. The work uses a comprehensive dataset for training and model evaluation which achieves a significant accuracy of 82.50%. The proposed methodology shows efficient results when measured against state-of-the-art methods for cultural event recognition over other datasets.