An extensive survey on traditional and deep learning-based face sketch synthesis models
- PDF / 404,438 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 68 Downloads / 216 Views
ORIGINAL RESEARCH
An extensive survey on traditional and deep learning-based face sketch synthesis models Narasimhula Balayesu1
•
Hemantha Kumar Kalluri1
Received: 11 February 2019 / Accepted: 4 November 2019 Bharati Vidyapeeth’s Institute of Computer Applications and Management 2019
Abstract In recent days, Face sketch synthesis (FSS) attracts various researchers for sketching the images to retrieve faces and in multimedia applications. The intention of FSS is to create a sketch for the image provided from a collection of sketch and photo images as the training set. Presently, the rise of deep learning (DL) models becomes useful in FSS because of its diverse benefits. As the FSS is employed in various applications, detailed experimentation to analyze the state of the art approaches methods is nontrivial. Though numerous FSS approaches are available, there is no review paper exist regarding the hierarchical classification of DL based FSS. Keeping this in mind, in this paper, we provide an extensive review of the available DL as well as conventional FSS techniques. We made a clear classification of the FSS techniques, and these are categorized into data-driven and model-driven methods. A comparative analysis of the reviewed techniques is made based on various aspects such as the objective, algorithms used, benefits, and performance measures. Keywords FSS Deep learning Face sketch Modeldriven Data-driven
& Narasimhula Balayesu [email protected] Hemantha Kumar Kalluri [email protected] 1
Department of Computer Science and Engineering, Vignan’s Foundation for Science, Technology & Research, Guntur, Andhra Pradesh, India
1 Introduction In recent days, machine learning (ML) concentrates on the process of defining input data and generalizing the learned patterns to employ in upcoming unknown data. The benefits of representing data have a greater influence on the outcome of the machine learners on the information. When the data is ineffectively represented, it has some negative impact on the results of the developed and complex machine learner, whereas the process of representing data in an efficient manner can results to the better outcome even for a simple machine learner. Therefore, feature engineering concentrates on the construction of feature and representation of data from original data [1]. It is a main component of ML, and it holds a major part of the work in the ML process and is naturally quite area-specific and includes significant input from humans. For instance, Histogram of Oriented Gradient (HOG) [2] and Scale-Invariant Feature Transform (SIFT) [3] are some of the famous feature engineering approaches mainly designed for computer vision applications. The process of presenting feature engineering in an automized way, and the general way can be a significant achievement in ML, and it enables the user to filter those features automatically without any human intervention. Deep Learning (DL) approaches are the significant methods in the area of study to the automatic filtering of co
Data Loading...
