Strengths and weaknesses of deep learning models for face recognition against image degradations

Grm, Klemen; Štruc, Vitomir; Artiges, Anais; Caron, Matthieu; Ekenel, Hazım Kemal

doi:10.1049/iet-bmt.2017.0083

Cited by 191 publications

(119 citation statements)

References 39 publications

Supporting

Mentioning

114

Contrasting

Unclassified

Order By: Relevance

“…The recognition models converge to the rank one recognition rate of 0.5138 (0.2974 † ) with 48 × 48px images, 0.7215 (0.4266 † ) with 96 × 96px images and 0.8569 (0.5713 † ) with 192×192px residual images on the training ( † validation) data. As expected, the performance decreases with a decreasing size of the residual images and is adversely affected by the lack of low-frequency information during training (see, e.g., [56] for the expected performance of SqueezeNet for face recognition). Nevertheless, the models contribute towards accurate and visually convincing SR results, as evidenced by the results in the next sections.…”

Section: Datasets and Model Trainingsupporting

confidence: 55%

Face Hallucination Using Cascaded Super-Resolution and Identity Priors

Grm

Scheirer

Štruc

2020

IEEE Trans. on Image Process.

Self Cite

View full text Add to dashboard Cite

Fig. 1. Sample face hallucination results generated with the proposed method.Abstract. In this paper we address the problem of hallucinating highresolution facial images from unaligned low-resolution inputs at high magnification factors. We approach the problem with convolutional neural networks (CNNs) and propose a novel (deep) face hallucination model that incorporates identity priors into the learning procedure. The model consists of two main parts: i) a cascaded super-resolution network that upscales the low-resolution images, and ii) an ensemble of face recognition models that act as identity priors for the super-resolution network during training. Different from competing super-resolution approaches that typically rely on a single model for upscaling (even with large magnification factors), our network uses a cascade of multiple SR models that progressively upscale the low-resolution images using steps of 2×. This characteristic allows us to apply supervision signals (target appearances) at different resolutions and incorporate identity constraints at multiplescales. Our model is able to upscale (very) low-resolution images captured in unconstrained conditions and produce visually convincing results. We rigorously evaluate the proposed model on a large datasets of facial images and report superior performance compared to the state-of-the-art.

show abstract

Section: Datasets and Model Trainingsupporting

confidence: 55%

Face Hallucination Using Cascaded Super-Resolution and Identity Priors

Grm

Scheirer

Štruc

2020

IEEE Trans. on Image Process.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Specifically, we include four recent state-of-the-art deep models in the experiments, i.e., AlexNet [ 59 ], VGG-Face [ 60 ], InceptionV3 [ 61 ] and SqueezeNet [ 62 ], and two techniques based on dense image descriptors, i.e., LBP (Local Binary Patterns) [ 63 ] and POEM (Patterns of Oriented Edge Magnitudes) [ 64 ]. We use pretrained publicly available deep models from [ 65 ] ) and the descriptor-based techniques from the AWE MATLAB toolbox [ 66 ] ) for our experiments. With all recognition methods, we simply extract features using the publicly available code and then use the computed feature vectors with the cosine similarity for recognition.…”

Section: Experiments and Resultsmentioning

confidence: 99%

k-Same-Net: k-Anonymity with Generative Deep Neural Networks for Face Deidentification

Meden

Emeršič

Štruc

et al. 2018

Entropy

Self Cite

View full text Add to dashboard Cite

; Tel.: +386-1-479-8245 † This paper is an extended version of our paper published in Meden B.; Emeršič Ž.; Štruc V.; Peer P.k-Same-Net: Neural-Network-Based Face Deidentification. In the Proceedings of the International Conference and Workshop on Bioinspired Intelligence (IWOBI), Funchal Madeira, Portugal, 10-12 July 2017.Received: 1 December 2017 ; Accepted: 9 January 2018; Published: 13 January 2018Abstract: Image and video data are today being shared between government entities and other relevant stakeholders on a regular basis and require careful handling of the personal information contained therein. A popular approach to ensure privacy protection in such data is the use of deidentification techniques, which aim at concealing the identity of individuals in the imagery while still preserving certain aspects of the data after deidentification. In this work, we propose a novel approach towards face deidentification, called k-Same-Net, which combines recent Generative Neural Networks (GNNs) with the well-known k-Anonymitymechanism and provides formal guarantees regarding privacy protection on a closed set of identities. Our GNN is able to generate synthetic surrogate face images for deidentification by seamlessly combining features of identities used to train the GNN model. Furthermore, it allows us to control the image-generation process with a small set of appearance-related parameters that can be used to alter specific aspects (e.g., facial expressions, age, gender) of the synthesized surrogate images. We demonstrate the feasibility of k-Same-Net in comprehensive experiments on the XM2VTS and CK+ datasets. We evaluate the efficacy of the proposed approach through reidentification experiments with recent recognition models and compare our results with competing deidentification techniques from the literature. We also present facial expression recognition experiments to demonstrate the utility-preservation capabilities of k-Same-Net.Our experimental results suggest that k-Same-Net is a viable option for facial deidentification that exhibits several desirable characteristics when compared to existing solutions in this area.

show abstract

“…However, after its creation, CNN's most powerful were presented, and at a specific instant, the creation of these powerful architectures stagnated. CNN's main problem is the large number of data to carry out the training process [9]. From this, techniques were proposed with the use of CNN's without performing all the training of the network.…”

Section: Related Workmentioning

confidence: 99%

Entropy Projection Curved Gabor with Random Forest and SVM for Face Recognition

Junior

Vogado

Rabelo

et al. 2019

Advances in Visual Computing

View full text Add to dashboard Cite

In this work, we propose a workflow for face recognition under occlusion using the entropy projection from the curved Gabor filter, and create a representative and compact features vector that describes a face. Despite the reduced vector obtained by the entropy projection, it still presents opportunity for further dimensionality reduction. Therefore, we use a Random Forest classifier as an attribute selector, providing a 97% reduction of the original vector while keeping suitable accuracy. A set of experiments using three public image databases: AR Face, Extended Yale B with occlusion and FERET illustrates the proposed methodology, evaluated using the SVM classifier. The results obtained in the experiments show promising results when compared to the available approaches in the literature, obtaining 98.05% accuracy for the complete AR Face, 97.26% for FERET and 81.66% with Yale with 50% occlusion.

show abstract

Strengths and weaknesses of deep learning models for face recognition against image degradations

Cited by 191 publications

References 39 publications

Face Hallucination Using Cascaded Super-Resolution and Identity Priors

Face Hallucination Using Cascaded Super-Resolution and Identity Priors

k-Same-Net: k-Anonymity with Generative Deep Neural Networks for Face Deidentification

Entropy Projection Curved Gabor with Random Forest and SVM for Face Recognition

Contact Info

Product

Resources

About