Accurate Object Detection and Semantic Segmentation using Gaussian Mixture Model and CNN

Abstract: Semantic segmentation and object detection are two most common tasks in the field of image processing, pattern recognition and classification. This paper presents a two stage procedure to perform these two tasks. The proposed work uses the Gaussian mixture model for image segmentation and identifies the segments by optimally searching the possible Gaussian distribution inside the image histogram. The optimal partition searching procedure uses the genetic algorithm. For the object detection, we apply the Convolution Neural Network (CNN) to extract the features of each segments and then apply them to pre-trained Support Vector Machine (SVM) to identify the object the segment belongs to. Finally the proposed system is developed using Matlab computational software and tested with different types of image datasets. The experimental results demonstrate encouraging performance of the proposed technique for both object detection and semantic segmentation tasks.

Keywords: Semantic segmentation, object detection, Gaussian mixture model, Genetic Algorithm, Support Vector Machine.