Entity Recognition in Indian Sculpture using CLAHE and machine learning

Sculpture recognition is one of the most challenging problems in the image classification field due to the high variations in the design of various sculptures. In order to classify the Indian entity's sculpture, we require images from multiple perspectives with different orientations of the structure. This research conducts a comparative study by combining various algorithms for the purpose of sculpture recognition based on their features. The SIFT (Scale Invariant Feature Transform) algorithm was used to find descriptors for the key points detected and it was paired with various classifiers (K-Nearest Neighbors, Support Vector Machine, Artificial Neural Network) by using the “Min key”, “Max key padding”, “Mean key padding”, “Median key padding” and “Mode key padding” approach for efficiency testing purposes. CNNs (Convolutional Neural Networks) were also tested for the same. The models were trained on various representations of different Indian sculptures, gathered from various sources, signifying our cultural diversity. Experiments were carried out on the manually acquired data set that consists of 15 different sculpture classes, where each sculpture class consists of 150 images for training and 20 for testing. An attempt was also made to increase the efficiency of these models by the application of CLAHE (Contrast Limited Adaptive Histogram Equalization). The experiments showed the performance of these models when they were trained on various representations of sculpture images. For 15 different sculpture classes, the maximum accuracy achieved was a respectable 70.66% utilizing the CLAHE along with the CNN model. However, the accuracy values of non-CNN-based approaches were substandard.