Red Green Blue Depth Image Classification Using Pre-Trained Deep Convolutional Neural Network
- PDF / 1,026,775 Bytes
- 9 Pages / 612 x 792 pts (letter) Page_size
- 96 Downloads / 175 Views
Red Green Blue Depth Image Classification Using Pre-Trained Deep Convolutional Neural Network N. Kumara,*, N. Kaura,**, and D. Guptaa,*** a
University Institute of Engineering and Technology, Panjab University, Chandigarh, India * e-mail: [email protected] ** e-mail: [email protected] *** e-mail: [email protected]
Abstract—Image Classification is one of the eminent challenges in the field of computer vision, and it also acts as a foundation for other tasks such as image captioning, object detection, image coloring, etc. The Convolutional Neural Networks (CNN) techniques have the potency to accomplish image classification for a variety of datasets. With the advancements in technologies, cameras are capturing high-level information such as depth. Therefore, it is essential to incorporate depth information into CNN to provide a better experience of image classification. In this paper, an attempt is made to adapt pre-trained GoogLeNet on Washington RGB-D (RGB-Depth) dataset. Moreover, GoogLeNet is evaluated on depth data that has provided reasonable classification rate on RGB-D dataset. In addition, the paper works on analyzing the impact of pre-processing or resizing of images and batch size on classification accuracy of the model. Keywords: Image Classification, Washington Red Green Blue Depth dataset, Computer Vision, Knowledge transfer DOI: 10.1134/S1054661820030153
1. INTRODUCTION Computer Vision (CV) [1] is a field of computer science and technology that tries to develop techniques to help the machines to view and understand the content or information of images as well as videos. Generally, CV is used to construct an artificial system which extracts data from the images or videos. It has been noticed that the use of Artificial Intelligence (Machine Learning or Deep Learning) in CV has provided lots of solutions to the problems like image classification, object detection and face recognition. Deep Learning (DL) has proved that it can attain state-ofthe-art results on these challenging problems. Deep Learning [2] is a sub-field in machine learning that is based on various layers of non-linear signal processing used for both supervised and non-supervised feature extraction for pattern analysis. The research techniques in the field of DL have influenced the broad range of information and signal processing tasks. Furthermore, the popular DL based CNN approach has effectively dealt with the signal processing tasks and problems in CV viz. image classification, object detection and face recognition, as well as opened the scope of AI to a greater extent. During the few past years, most of the techniques that have emerged for the RGB-D (RGB-Depth)
Received August 6, 2019; revised August 6, 2019; accepted February 20, 2020
classification relied on descriptor-based approaches such as SIFT (Scale Invariant Feature Transform) [3] SURF (Speeded Up Robust Features) [4] in collaboration with shape features. The LeNet model introduced by Lecun et al. [5] was one of the first CNN models popularly known as root o
Data Loading...
