Comparative Analysis of RCNN and Faster R-CNN for Text Detection in a Poor Quality Topographic Map

Abstract

Topographic Map (TM) highlights essential morphological features. Some notable features include contours, elevation values, etc. It may be either distinct or overlapped with other features. All these attributes are annotations written using distinct color codes. These features along with their associated attributes is used for various morphological analyses. The process of extracting morphological features along with attributes from map is referred as digitization. Digitization is done manually or with an abled automatic computational process. The manual process is tedious and time-consuming whereas automated process takes comparatively lesser time but generates quality results. With advanced learning techniques, quality and integrity of result have considerably increased. CNN adopts traditional MLP where convolution layers replaces intermediate hidden layers in the network. Convolutional layers build a versatile mechanism to enable the detector to identify the features. This paper presents comparison of RCNN and Faster RCNN model result to detect and locate text present in the TM.