Cell Detection in Tabular data

Abstract

The primary goal of our project is to create a non - deep learning solution for effectively segmenting cells within tabular data, accommodating tables with or without gridlines.

We have devised an algorithm based on K-Means Clustering to facilitate cell segmentation within tables, irrespective of the presence of gridlines. Our approach involves identifying clusters of characters, often representing words or numbers, and subsequently calculating their centres of mass. We create distinct arrays for the x and y coordinates of these centres. Employing K-Means clustering separately on x coordinates and y coordinates of centres, we determine the optimal number of clusters, denoted as 'k,' from 1 to a predefined maximum value ('max_k') using a novel method for selecting the most suitable 'k', as the existing methods yielded unsatisfactory results. Subsequently, we discern rows and columns separately by employing K-Means clustering with the determined 'k' and identify individual cells through the intersection of these rows and columns.

In addition, we have developed an alternative algorithm tailored for tables containing gridlines. In this scenario, we use canny edge detection and hough transform to detect lines, followed by the identification of intersection points. We use intersection points to detect gridlines. Using these detected gridlines, we reconstruct the table structure.