Paper Title
GABOR FILTER-BASED DRAWING IMAGE INSTANCE SEGMENTATION ENHANCEMENT USING MASKED RCNN ON LIMITED DATASET

Abstract
Abstract - This paper presents the implementation of a Gabor filter to show superior improvement in terms of instance segmentation on a limited drawing image dataset. Two versions of the Mask Recurrent Neural Network (MRCNN) model were used in the experiment: the original architecture of MRCNN and a MRCNN that utilized Gabor Filter processing on the image prior to the image segmentation. This aims to address the challenges on image segmentation for datasets with non obvious properties like freehand drawings and the use of limited datasets. This paper envisions further use of the improved model in drawing analysis as a tool for psychological assessment. The model yielded an overall accuracy of 91%, outperforming the accuracy of the original benchmark model of 87%. The differences of the stratified k-fold validations performed is statistically significant at p-value of 0.0064. Keywords - Gabor Filter, Image Segmentation, Drawing Object Classification, MRCNN, Limited Dataset.