CSI6209-A3-Part 1 (Implementation-Report)-Rubric [Version 2021-07-15 0903]
CSI6209-A3-Part 1 (Implementation-Report)-Rubric [Version 2021-07-15 0903]
CSI6209-A3-Part 1 (Implementation-Report)-Rubric [Version 2021-07-15 0903]
Criteria Ratings Pts
This criterion is linked to a learning outcomeReport-Abstract 3Pts
Excellent
Includes all of the following excellently: Engaiging starting sentence, brief description of approach, performance report 2.4Pts
Very Good
Includes all but not in the best manner: Misses one of the following: Engaiging starting sentence, brief description of approach, performance report 1.8Pts
Good
Misses one of the following: Engaiging starting sentence, brief description of approach, performance report 1.2Pts
Poor
Misses two of the following: Engaiging starting sentence, brief description of approach, performance report 0.3Pts
Very Poor
Some attempts have been made
3pts
This criterion is linked to a learning outcomeReport- Introduction 4.5Pts
Excellent
Gives an excellent overview dragging interest of the reader, cites at least 5 related references, describe chosen algorithm(s) in short and motivation for the choice, mention organization of the report 3.6Pts
Very Good
Very nice short introduction to the approach. 2.25Pts
Good
Covers all the necessary components 1.35Pts
Poor
Misses one or two components 0.45Pts
Very Poor
Misses one or two major components
4.5pts
This criterion is linked to a learning outcomeReport-Research Approach 10.5Pts
Excellent
Gives an excellent description with: an overall block diagram/flowchart, a description of each block/component, a good number of clear and correct illustrations (e.g. equations/figures/pseudocode, etc), the inclusion of code (or chronological screenshots of tools used) in the appendix or separately in a zipped file. 8.4Pts
Very Good
Include all but not in correctly and clearly 6.3Pts
Good
Misses few items 3.15Pts
Poor
Misses most of the items 0Pts
Very Poor
Just copy and paste of other peoples work
10.5pts
This criterion is linked to a learning outcomeReport-Performance Evaluation 7.5Pts
Excellent
Compare the performance of the chosen algorithm with respect to at least another algorithm with very clear and correct plots/figures. All executable/runnable code files and test datasets are uploaded as a zipped file. 6Pts
Very Good
Include all but not in correctly and clearly 4.5Pts
Good
Misses few items. 2.25Pts
Poor
Misses most of the items 0Pts
Very Poor
No attempts have been made
7.5pts
This criterion is linked to a learning outcomeReport- Conclusion 3Pts
Excellent
Summarise very clearly what was done and learned, mentions prospects and limitations and possible improvements. 2.4Pts
Very Good
Includes a very concise summary of what was done and learned. 1.8Pts
Good
Misses either performance or learning summary 1.2Pts
Poor
Incomplete conclusion, missing summary of both learning and performance. 0Pts
Very Poor
No conclusions present.
3pts
This criterion is linked to a learning outcomeReport-References 1.5Pts
Excellent
Five or more relevant references are used and correctly formated1.2Pts
Very Good
Five or more relevant references are used, but not correctly formated0.9Pts
Good
Less than five relevant references are used and may not correctly formated0.6Pts
Poor
Less than three relevant references are used and may not correctly formated0.3Pts
Very Poor
Less than two relevant references are used but incorrectly formated1.5pts
Total points:30
Title: A Comprehensive Survey on Automatic Human Recognition Using 2D and 3D Ear and Face Images
Contents
TOC o "1-3" h z u Introduction PAGEREF _Toc145259442 h 3Classification or Taxonomy: PAGEREF _Toc145259443 h 3Description of Different Approaches PAGEREF _Toc145259444 h 5An explanation of 2D face recognition techniques PAGEREF _Toc145259445 h 5An explanation of 2D ear recognition methods PAGEREF _Toc145259446 h 6An explanation of 3D face recognition techniques PAGEREF _Toc145259447 h 8Detailed Description of 3D Ear Recognition Methods PAGEREF _Toc145259448 h 9Challenges and Future Research PAGEREF _Toc145259449 h 11Conclusion PAGEREF _Toc145259450 h 12References PAGEREF _Toc145259451 h 13Research Plan for Assessment-2 Part-2: PAGEREF _Toc145259452 h 14
IntroductionThe report's fundamental backdrop is provided by the introduction and background portion of a thorough assessment of automated person recognition utilising 2D and 3D ear and facial pictures. This section begins with a definition of the issue, which centres on the creation and improvement of recognition systems based on ear and face biometrics. Due to its applications in security, identity verification, and access control, the recognition of persons based on face and ear traits has greatly increased in significance.
The importance of this issue is shown by the fact that it affects a number of industries, including personal device security, healthcare, and law enforcement. Secure access to both physical and digital locations and the protection of sensitive information depend on accurate and effective human identification. This introduction also covers essential ideas and terms related to face and auditory recognition methods to aid in a thorough comprehension of the next parts. For the benefit of readers, especially those who are unfamiliar with the subject, definitions of important concepts like "deep learning," "biometric authentication," and "3D face recognition" are provided.
The report provides a road map for the reader by outlining its organisational structure. The parts that follow will go into great length on the pertinent literature, classifying methods, going through difficulties, and suggesting new paths for future study in the area of automatic human recognition utilising 2D and 3D ear and face pictures. This structure guarantees a methodical investigation of the issue, improving the reader's comprehension of the material. To support the discussion and analysis of the chosen issue, a number of sources, including Kong et al. (2023), Kamboj et al. (2022), and others, will be cited throughout the study (Kong et al., 2023; Kamboj et al., 2022).
In conclusion, this introduction demonstrates the importance of the issue, presents crucial ideas, and gives the rest of the study a clear framework, paving the way for a thorough investigation of automatic human recognition utilising 2D and 3D ear and face pictures.
Classification or Taxonomy:According to the artificial intelligence (AI) methodologies they use for automatic human detection utilising 2D and 3D ear and face pictures, the different approaches and methods assessed in this research are categorised in the survey's Classification/Taxonomy section. For readers to receive a well-structured and organised overview of the area, this categorisation is crucial.
The surveyed approaches will be divided into separate groups based on the AI techniques used to achieve this categorization. Notably, some strategies could use conventional machine learning techniques while others might make use of cutting-edge deep learning technologies. This classification enables readers to recognise the variety of AI approaches applied in this field and to acquire understanding of the advantages and disadvantages of each category.
For instance, Kong et al. (2023) introduced a deep learning-based 3D face recognition system, highlighting the value of convolutional neural networks (CNNs) and the Laplacian pyramid (Kong et al., 2023). However, Kamboj et al. (2022) completed a thorough assessment that includes deep learning-based methods for ear biometric detection, demonstrating the prominence of deep learning in this sector.
Traditional machine learning techniques that have been applied in prior methods, such as principal component analysis (PCA) or support vector machines (SVMs), may fall under other categories. This section gives readers a clear picture of the methodological landscape in the field of automatic human recognition by categorising the studied methods based on AI techniques. It also lays the groundwork for the sections to follow, which will investigate each approach type in detail, including its techniques, data sources, and performance measures.
We will categorise the many techniques used in this field in order to offer a comprehensive overview of the research landscape in automatic human recognition utilising 2D and 3D ear and facial pictures. We shall be able to classify and comprehend the numerous AI methods used for this aim with the aid of this categorization.
Figure 1: Taxonomy of Automatic Human Recognition Approaches
2D Face Recognition 2D Ear Recognition
3D Face Recognition
3D Ear Recognition
Traditional Methods (e.g., Eigenfaces, Fisherfaces) Shape-based Methods
Depth-based Methods
Surface-based Methods
Deep Learning-based Methods (e.g., Convolutional Neural Networks) Texture-based Methods Mesh-based Methods
Volumetric Methods
We develop a systematic framework to examine the various strategies used in automatic human recognition by classifying the approaches into four groups. The methodology, datasets, and performance assessments for each category will be thoroughly discussed in the sections that follow.
This taxonomy not only clarifies things but also acts as the basis for a thorough study of the methodologies that have been examined. It improves readers' understanding of this dynamic topic by letting them browse and comprehend the AI methods utilised in many aspects of autonomous human identification.
Description of Different ApproachesAn explanation of 2D face recognition techniquesThe field of automated human recognition has seen a lot of research in 2D face recognition. It comprises a number of strategies, each with particular characteristics, benefits, and drawbacks. We outline the main strategies in this area in this section.
Eigenfaces:
Principal Component Analysis (PCA) is the foundation of the classic approach known as Eigenfaces. The main components (or eigenfaces) that are taken from a collection of face photos and combined linearly to represent faces.
Benefits: Eigenfaces was a popular choice for early face recognition systems since it is straightforward and computationally efficient.
Its sensitivity to changes in illumination is one of its main drawbacks. It could have trouble identifying faces in different lighting situations.
CNNs (convolutional neural networks)
Key characteristics: CNNs, a subset of deep learning techniques, have completely changed 2D face recognition. To automatically extract hierarchical characteristics from face photos, they use numerous layers of convolutional and pooling processes.
Benefits: CNNs can capture intricate patterns and changes in facial characteristics and have great accuracy in face recognition applications.
Limitations: Large datasets are frequently needed for training CNNs, although they are not always easily accessible. They might also be computationally demanding.
Fig.1 Face recognition methods. ( source: https://www.mdpi.com/1424-8220/20/2/342 )
Method Key Features Advantages Limitations
Eigenfaces Principal Component Analysis Simple and efficient Sensitive to lighting changes
CNNs Deep learning High accuracy Requires large datasets
Table 1: Summary of 2D Face Recognition Approaches
An explanation of 2D ear recognition methodsAutomatic person recognition systems must include 2D ear recognition, which focuses on the distinctive qualities of ear features. We give an overview of the most important methods in this area in this section.
Using EigenearsPrincipal Component Analysis (PCA) is the foundation of the Eigenears approach, which was developed as an inspiration for Eigenfaces. The primary components taken from a collection of ear pictures and combined into linear combinations to describe ear images are called eigenears.
Benefits: Eigenears provides a straightforward and computationally effective method for 2D ear detection. In situations where ear photos are well-lit and show little fluctuations, it may be useful.
Limitations: Like Eigenfaces, Eigenears may have trouble with ear photos with large changes in lighting and position.
Texture-based Approaches as a Method
Key characteristics: Texture-based techniques for 2D ear identification study texture patterns in captured ear pictures. For feature extraction, methods like Gabor filters and local binary patterns (LBP) are frequently used.
Advantages: Texture-based techniques are resistant to some fluctuations in illumination and position since they are able to capture detailed patterns in ear pictures.
Limitations: They can need a bigger dataset to generalise successfully, and the quality of the texture characteristics collected can have an impact on how well they function.
Fig.2 The suggested 2D-MBPCA algorithm's block diagram (Source: https://www.researchgate.net/publication/360055057_Multi-Band_PCA_Based_Ear_Recognition_Technique )
Method Key Features Advantages Limitations
EigenearsPrincipal Component Analysis Simple and efficient Sensitive to lighting changes
Texture-based Gabor Filters, Local Binary Patterns Texture pattern analysis Dataset size and feature quality
Table 2: 2D Ear Recognition Methods in Summary
An explanation of 3D face recognition techniquesDifferent techniques have arisen in the field of 3D face recognition, each with their own distinctive characteristics, benefits, and drawbacks. We give an overview of several important techniques under this category in this section, accompanied by pertinent research.
Depth-based Approaches as a Method
Key characteristics: According to Kong et al. (2023), depth-based approaches make use of depth data obtained by specialised sensors. They depict face features in three dimensions, making them resistant to changes in illumination.
Benefits: Depth-based methods rely on geometric elements rather than outward appearance, which makes them resistant to changes in illumination. They are therefore appropriate for use in practical settings.
Limitations: However, one significant flaw is that they depend heavily on specialised sensors, which may not always be feasible or easily accessible (Kong et al., 2023).
Mesh-based methods are used
Key characteristics: Mesh-based systems describe facial surfaces as mesh structures, which are frequently obtained by 3D scanning or modelling methods. For identification, they use the geometrical features of face meshes.
Benefits: Mesh-based techniques usually achieve high accuracy in 3D face recognition challenges because they can accurately capture the complex facial geometry (Chen & Wu, 2021).
Limitations: On the other hand, analysing complicated mesh data can be computationally demanding and may call for a significant amount of CPU power.
Fig.3 Flowchart for 3D point cloud facial reconstruction; blue blocks represent preprocessing processes, orange blocks represent registration steps, and green blocks represent denoising techniques. ( Source: https://www.mdpi.com/1424-8220/21/8/2587 )
Method Key Features Advantages Limitations
Depth-based Depth information Robust to lighting Requires specialized sensors (Kong et al., 2023)
Mesh-based Surface mesh representation High accuracy Complex data processing (Chen & Wu, 2021)
Table 3: Summary of 3D Face Recognition Approaches
Detailed Description of 3D Ear Recognition MethodsIn the field of automatic human recognition, 3D ear recognition is a specialised field that makes advantage of the particular characteristics of ear structures. An overview of the main methods in this area is given in this part, along with information on their special qualities, benefits, and drawbacks, all backed by pertinent research.
Approaches based on the surface
Key characteristics: Surface-based methods for 3D ear identification study the curvature and surface properties of ear components. These techniques frequently entail the 3D ear scans' curvature data extraction.
Benefits: Since surface-based approaches concentrate largely on the innate surface geometry of the ear, they are resilient to changes in position. They are therefore useful for identifying ears in various directions.
Limitations, however, include the complexity of data collecting since it can be difficult to produce reliable 3D ear scans and may need for specialised equipment (Li et al., 2018).
Volumetric approaches are used
Key characteristics: Voxel-based representations are used in volumetric techniques to portray ear components. By assembling voxels into a three-dimensional grid, they are able to record the complete ear volume.
Benefits: Volumetric techniques are renowned for their immunity to noise and their capacity to record accurate 3D data. This resilience is especially helpful in demanding and loud conditions.
Limitations: On the negative side, processing volumetric data computationally can be expensive and demand substantial computer resources for recognition tasks (Li et al., 2018).
Fig.4 VoxelNet Architecture Overview ( source: https://www.mdpi.com/1099-4300/25/4/635 )
Summary of 3D ear recognition techniques (Table 4)
Method Key Features Advantages Limitations
Surface-based Surface curvature analysis Robust to pose changes Data acquisition complexity (Li et al., 2018)
Volumetric Voxel-based representation Insensitive to noise High computational cost (Li et al., 2018)
Challenges and Future ResearchThe field of automatic human recognition, especially in the context of biometrics-based identification, medical imaging-based diagnosis, and object/anomaly detection, provides a wide range of difficulties and promising directions for further study. With the use of the knowledge gained from the literature, we address the problems frequently mentioned in research publications in this part and suggest prospective topics for further study.
Performance Differences Among Various Methods:
The diversity in performance among various AI-based recognition techniques is one of the persistent problems noted in the research. The selection of methods, dataset size and quality, and the presence of environmental elements like illumination and occlusions are factors that affect this difference. The creation of techniques capable of providing constant performance and resilient to these variances is necessary to address this difficulty.
Enhancing Robustness: Future research projects should concentrate on creating recognition techniques that are resilient to changes in stance, illumination, and occlusions. This might entail developing new algorithms that can adjust to shifting environmental conditions or using deep learning methods that can learn invariant representations.
Data augmentation: Access to data continues to be a major barrier, particularly for 3D data utilised in biometrics and object identification. Researchers might look for data augmentation approaches to lessen this restriction. Researchers may increase the diversity and volume of training data by creating synthetic data and supplementing existing datasets, which will eventually enhance the performance and generalisation of AI models.
Cross-Modal Recognition: Cross-modal recognition is a promising new field in which AI systems are able to match and recognise humans or things across several data modalities, such as matching 2D and 3D data. It may be possible to improve the precision and applicability of recognition algorithms in mixed-data contexts by looking into techniques for harmonising information across modalities.
Privacy and Security: As the use of biometric data for identification increases, privacy and security issues are of utmost importance. Future studies should examine privacy-preserving methods for biometric data to protect sensitive information. Furthermore, it is vital to investigate ways to strengthen security against spoofing attacks, in which hostile actors try to trick identification systems.
The subject of automated human identification has a great opportunity to improve by tackling these issues and adopting new research approaches, leading to more precise, reliable, and safe AI systems for a variety of applications. These initiatives will surely help the fields of artificial intelligence and biometrics continue to develop and innovate.
ConclusionUsing photos of the face and ears in 2D and 3D, this survey has given a thorough review of automatic human recognition. We divided the methods into categories, enumerated their salient characteristics, and talked about problems and possible future lines of inquiry. Despite the fact that a lot has been accomplished, there are still issues to be resolved, making this an attractive subject for further study.
Limitations:
It's crucial to be aware of this survey's limitations. Despite our best efforts to give a thorough overview, fresh approaches and discoveries may arise that fall beyond the purview of this analysis due to the fast-moving pace of AI research. Due to the variability of the available research, the degree of coverage for each technique may also differ.
Next Steps:
This survey's recommendations for difficulties and future research projects set the stage for exciting new work in autonomous human recognition:
Advanced Robustness: Research efforts should concentrate on creating even more resilient recognition systems that can resist a larger range of environmental difficulties.
Enhanced Data Augmentation: Research into data augmentation methods should continue, especially for 3D data, since this will help us get over data constraints.
Advancements in cross-modal recognition techniques will make recognition systems more adaptable.
Innovations in privacy and security: It is crucial to conduct more research to provide privacy-preserving methods and stronger defences against spoofing assaults.
In conclusion, despite the impressive advancements gained, the area of automatic human recognition utilising 2D and 3D photographs of the ears and faces is still young and active, with plenty of room for innovation and advancement. We anticipate seeing even more potent and robust identification systems in the near future as AI approaches continue to progress, tackling real-world difficulties and boosting security and convenience across multiple areas.
References1. Kong, W., You, Z., & Lv, X. (2023). 3D face recognition algorithm based on deep Laplacian pyramid under the normalization of epidemic control. Journal of Computer Communications, 199, 30-41. https://www.elsevier.com/locate/comcom
2. Kamboj, A., Rani, R., & Nigam, A. (2022). A comprehensive survey and deep learning-based approach for human recognition using ear biometric. The Visual Computer, 38(10), 23832416. HYPERLINK "https://doi.org/10.1007/s00371-021-02119-0"https://doi.org/10.1007/s00371-021-02119-0
3. Dirin, A., Delbiaggio, N., & Kauttonen, J. (2020). Comparisons of Facial Recognition Algorithms Through a Case Study Application. International Journal of Interactive Mobile Technologies (iJIM), 14(14), pp. 121 133. https://doi.org/10.3991/ijim.v14i14.14997
4. Li, J., Qiu, T., Wen, C., Xie, K., & Wen, F.-Q. (2018). Robust Face Recognition Using the Deep C2D-CNN Model Based on Decision-Level Fusion. Sensors, 18(7), 2080. https://doi.org/10.3390/s18072080
5. Mursal, M., & Islam, S. M. S. (Senior Member, IEEE). (2021). Deep Learning for 3D Ear Detection: A Complete Pipeline from Data Generation to Segmentation. IEEE Access. Advance online publication. https://doi.org/10.1109/ACCESS.2021.3129507Research Plan for Assessment-2 Part-2:Automatic Human Recognition using 2D and 3D ear and face images
Aims and Objectives:
Hybrid Recognition System Development: Using the advantages of both modalities, develop an integrated 2D and 3D ear and face recognition system.
Performance Assessment: Perform extensive testing to assess the reliability, robustness, and computational effectiveness of the recognition system.
Analyse performance of the hybrid system in comparison to separate 2D and 3D recognition techniques, emphasising gains and trade-offs.
Research Techniques:
1. Data gathering: Compile large datasets of 2D face, 2D ear, 3D face, and 3D ear photos, making sure that the subjects, lighting, and position circumstances are varied.
2. Preprocessing: To get the datasets ready for recognition tasks, use preprocessing techniques including data alignment, quality improvement, and normalisation.
3. Hybrid Recognition Model: Create a hybrid deep learning model that blends mesh-based techniques for 3D data with Convolutional Neural Networks (CNNs) for 2D data. Implement transfer learning to achieve the best results.
4. Evaluation: Using cross-validation trials to measure accuracy, precision, recall, F1-score, and processing time, thoroughly evaluate the performance of the hybrid recognition system.
5. Comparative Analysis: Evaluate the individual recognition abilities of 2D and 3D data, as well as the advancements made by the hybrid system.
Set of Data to Be Used:
Use open-access datasets like LFW for 2D face pictures, the University of Bradford Ear Database for 3D ear scans, and 3D facial expression datasets for 3D face photos. The wide range of datasets chosen will make thorough analyses easier.
Expected Results:
Through the creation of a hybrid recognition system, this research plan intends to enhance automated human recognition utilising 2D and 3D ear and face pictures. The review and comparison study will shed light on the system's potential uses in a variety of industries,including security, healthcare, and more, while also offering insightful information about its capabilities.
