Face Sketch Creation and Recognition

Abstract

This paper presents a wisdom mapping approach to dissect the thematic elaboration of face recognition exploration. For this reason, different bibliometric tools are combined (Performance analysis, wisdom mapping and Coword analysis) to identify the most important, productive and loftiest- -impact sub fields. also, different visualization tools are used to display a graphical vision of the face recognition field to determine the thematic disciplines and their evolutionary geste. . Eventually, this study proposes the most applicable lines of exploration for the face recognition field. Findings indicate a huge increase in face recognition exploration since 2014. Mixed approaches revealed a great interest compared to original and global approaches. In terms of algorithms, the use of deep literacy styles is the new trend. On the other hand, the illumination variation impact on face recognition algorithms\' performances is currently, the most important and impacting challenge for the face recognition field.

Introduction

I. INTRODUCTION

The human face is crucial for social interaction and communication, aiding in recognizing individuals in various settings. From interpersonal connections to technological applications, face recognition enhances interactions and security. Traditionally, it has been used in access control, security systems, and surveillance, optimizing workflow efficiency and safety protocols. However, modern technology demands refined methodologies to meet current needs. Recent advancements in face recognition technology have led to sophisticated algorithms improving accuracy and efficiency. Two primary modes, face verification and face identification, underscore its diverse applications. Face verification determines if two images depict the same individual, while face identification identifies a person from a pool of candidates, highlighting the complexity of implementation.

Our research addresses challenges in face recognition by proposing a hybrid neural network solution. Integrating diverse computational techniques inspired by human perception, our approach combines local image sampling, a self-organizing map (SOM) neural network, and a convolutional neural network (CNN) for robustness, dimensionality reduction, and partial invariance to factors like translation, rotation, scale, and deformation.

Additionally, we explore alternative techniques such as the Karhunen–Loeve (KL) transform and a multilayer perceptron (MLP) to compare their efficacy with our hybrid model. Through experimentation and analysis, we demonstrate superior performance, especially in scenarios with limited training data and variability in facial expressions, poses, and details. Leveraging a comprehensive database of 400 images from 40 individuals, our solution proves scalable and adaptable across real-world applications. Beyond performance metrics, we consider computational complexity and strategies for accommodating new classes within the trained recognizer. By examining the thematic evolution of face recognition research and identifying emerging trends, we contribute to ongoing discussions in this dynamic field. Ultimately, our research aims to advance face recognition technology, enhancing security measures, human-computer interaction, and criminal identification processes.

A. Classical Face Sketch Creation

Classical face creation encompasses a wide array of techniques, ranging from traditional methods to cutting-edge technologies. One common approach involves utilizing specialized software platforms tailored for crafting digital portraits or composite sketches. These platforms offer a plethora of tools enabling users to manipulate facial features, tweak proportions, and add intricate details like hair, eyes, and skin texture. Whether it's for artistic expression or forensic investigation, these digital tools empower artists and investigators to produce realistic depictions of faces, drawing inspiration from their imagination or reference photographs.

Furthermore, classical face creation often entails leveraging pre-existing sketches or templates as a starting point for further customization. Forensic artists, for instance, may commence with a rudimentary outline or composite sketch provided by a witness, refining it with additional details or descriptions to achieve a more accurate portrayal. Similarly, artists across various domains may utilize stock images, historical portraits, or artistic references as foundational elements for their creations. They adapt and modify these references to align with their specific requirements or artistic vision, resulting in unique and compelling representations.

In addition to digital tools and pre-existing references, classical face creation can also involve traditional techniques such as sketching by hand or sculpting clay models. These traditional methods offer artists a hands-on approach to capturing the intricacies of facial anatomy and expression, allowing for a deeper level of artistic interpretation and creativity.

Moreover, advancements in technology have expanded the repertoire of classical face creation techniques. Techniques like 3D modeling and rendering enable artists to sculpt lifelike faces in virtual environments, adding a new dimension to the creative process. Furthermore, machine learning algorithms can analyze vast datasets of facial images to generate photorealistic portraits or predict facial features based on given parameters, pushing the boundaries of what's possible in face creation.

B. Feature Extraction

In the realm of face recognition, feature extraction refers to the process of identifying and isolating crucial facial attributes or characteristics from images or datasets. These attributes act as distinct identifiers, enabling precise differentiation between individual faces and forming the foundation for accurate classification and identification. Essentially, feature extraction entails converting raw visual data into a more condensed and representative format, facilitating efficient analysis and comparison.

A diverse array of techniques is employed for feature extraction in face recognition systems, each targeting specific aspects of facial appearance and structure. One prevalent method involves extracting geometric features, which encompass the positions and proportions of critical facial landmarks like the eyes, nose, mouth, and chin. By quantifying the spatial relationships between these landmarks, a unique geometric profile is generated for each face. This profile serves as a basis for matching and recognition tasks, contributing to the robustness and accuracy of the overall system.

In addition to geometric features, other types of facial attributes can also be extracted during the feature extraction process. These attributes may include texture patterns, such as skin texture and wrinkles, as well as statistical features like pixel intensity distributions. By combining multiple types of features, face recognition systems can achieve a more comprehensive understanding of facial characteristics, enhancing their ability to discriminate between individuals accurately

Furthermore, advancements in deep learning and convolutional neural networks (CNNs) have revolutionized the field of feature extraction in face recognition. CNNs can automatically learn hierarchical representations of facial features directly from raw image data, eliminating the need for manual feature engineering. This data-driven approach enables the extraction of complex and abstract features, leading to significant improvements in recognition accuracy and robustness.

C. Classical Face Detection

Early face detection methods, often referred to as classical face detection, mark the genesis of contemporary techniques employed to detect and pinpoint human faces within images. These foundational approaches have shaped the landscape of modern face detection technologies, serving as the cornerstone for a myriad of applications prevalent today.

Initially, classical face detection algorithms relied heavily on manually crafted features and conventional deep learning methodologies such as Convolutional Neural Networks (CNNs). Their primary objective was to discern prominent facial attributes like eyes, nose, and mouth by leveraging pre-established patterns and templates.

The evolution of classical face detection paved the path for more advanced methodologies, particularly those rooted in deep learning frameworks like CNNs. These state-of-theart techniques have revolutionized face detection capabilities, substantially enhancing accuracy and resilience. Consequently, they have become indispensable tools across a spectrum of applications, spanning from security systems to social media platforms.

III. LITERATURE REVIEW

Various studies have delved into different techniques for creating and recognizing facial sketches, shedding light on the challenges and advancements within the field. For instance, one study introduced an independent application tailored for constructing and identifying facial composites. Initially, the conventional approach proved cumbersome and perplexing. However, a novel method allowed victims to select faces resembling the suspect from a pool of options. Subsequently, the system amalgamated these selections to automatically generate the criminal's facial composite, yielding promising outcomes in accurately identifying composite faces. Another notable approach presented a recognition method for synthesizing photo-sketches utilizing a Multiscale Markov Random Field Model. This method facilitated the synthesis of sketches from photos and vice versa, enabling database searches for pertinent matches. By segmenting facial sketches into patches and training the model to minimize disparities between photos and sketches, the efficacy of the recognition model was augmented.

Conclusion

The project \"Forensic Face Sketch Construction and Recognition\" has been meticulously designed, developed, and rigorously tested to cater to real-world scenarios, prioritizing security, privacy, and accuracy at every stage. From the initial splash screen to the final data retrieval phase, the platform has been engineered with these key factors in mind. Significantly, the platform has demonstrated exceptional security measures by implementing strict protocols. For instance, it verifies the user\'s credentials by matching the MAC Address and IP Address during the loading process, effectively blocking unauthorized access. Additionally, the OTP system adds an extra layer of security by preventing the reuse of previously generated OTPs and generating new ones with each page reload or login attempt. Moreover, the platform has exhibited impressive accuracy and speed in both face sketch construction and recognition tasks. During extensive testing with diverse test cases, scenarios, and datasets, it achieved an average accuracy rate of over 90%, coupled with a confidence level of 100%. These results align well with existing studies in the field, indicating commendable performance.

References

[1] Hamed Kiani Galoogahi and Terence Sim, “Face Sketch Recognition By Local Radon Binary Pattern: LRBP”, 19th IEEE International Conference on Image Processing, 2012. [2] Charlie Frowd, Anna Petkovic, Kamran Nawaz and Yasmeen Bashir https://ieeexplore.ieee.org/document/5376857 [3] P. Yuen and C. Man: https://ieeexplore.ieee.org/document/4244565. [4] N. Wang, X. Gao, and J. Li, “Random sampling for fast face sketch synthesis,” Pattern Recognit., vol. 76, pp. 215– [5] K. Elissa, N.Wang, X.Gao, and J. Li, “Random sampling for fast face sketch synthesis,” Pattern Recognit., vol. 76, pp.215-227,2018. [6] H. Han, B. Klare, K. Bonnen, and A. Jain, “Matching composite sketches to face photos: A component-based approach,” IEEE Trans. on Information Forensics and Security, vol. 8, pp. 191–204, January 2013. [7] K. Fukunaga, Introduction to Statistical Pattern Recognition, 2nd ed.Boston, MA: Academic, 1990 [8] T. Kohonen, “The self-organizing map,” Proc. IEEE, vol. 78, pp.1464–1480, 1990. [9] N.Wang,X.Gao,and J.Li,“Randomsamplingforfast face sketch synthesis,” Pattern Recognit., vol. 76,pp.215–227,2018. [10] X. Wu, L. Song, R. He, and T. Tan, ‘‘Coupled deep learning for heterogeneous face recognition,’’ in Proc. AAAI Conf. Artif. Intell., 2018, vol. 32,no. 1, pp. 1–

Copyright

Copyright © 2024 Prof. Anoop Kushwaha, Sanket Kale, Pramod Rumale, Abhishek Gavhane, Mayur Shinde. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.