Design and Implementation of 2D Object Similarity Detection using Hu’s Moments

Object similarity detection is the process of searching and identifying similarities between the target image and the test images. It is crucial for tasks like object recognition and automated inspection systems. This paper presents a novel methodology for object similarity detection using template matching. It employs Hu's moments renowned for their invariance properties. The first stage involves preprocessing the images through grayscale conversion and resizing to standardize their dimensions. Raw moments are computed to capture the spatial distribution of pixel intensities within both the target and the test images. Centroids are then determined to characterize the centroidal location of pixel intensities. Following this, moment normalization is conducted to ensure the robustness of features against translation and rotation variations. Template matching is accomplished through a sliding window technique, wherein local regions of the test image are compared with the target object using the extracted moments. Polynomial regression is performed to calculate the acceptable deviation range between Hu’s moments of the target image and the test images. The identified similarities serve as indicators of object correspondences, thus demonstrating the efficacy of the proposed method across diverse rotations and translations.