Face recognition with Patch-based Local Walsh Transform

Uzun‐Per, Meryem; Gokmen, M.

doi:10.1016/j.image.2017.11.003

Cited by 12 publications

(4 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The use of distillation-based learning methods [40,41] to estimate a simpler model with similar performance is another line to follow. Besides, we consider the integration of a previous speaker identification module [42]. Thus, unlike our experiments where a perfect speaker classifier was assumed, we will be able to develop a system aimed at a realistic application.…”

Section: Discussionmentioning

confidence: 99%

Speaker-Adapted End-to-End Visual Speech Recognition for Continuous Spanish

Gimeno-Gomez¹,

Hinarejos²

2022

IberSPEECH 2022

View full text Add to dashboard Cite

Different studies have shown the importance of visual cues throughout the speech perception process. In fact, the development of audiovisual approaches has led to advances in the field of speech technologies. However, although noticeable results have recently been achieved, visual speech recognition remains an open research problem. It is a task in which, by dispensing with the auditory sense, challenges such as visual ambiguities and the complexity of modeling silence must be faced. Nonetheless, some of these challenges can be alleviated when the problem is approached from a speaker-dependent perspective. Thus, this paper studies, using the Spanish LIP-RTVE database, how the estimation of specialized end-to-end systems for a specific person could affect the quality of speech recognition. First, different adaptation strategies based on the fine-tuning technique were proposed. Then, a pre-trained CTC/Attention architecture was used as a baseline throughout our experiments. Our findings showed that a two-step finetuning process, where the VSR system is first adapted to the task domain, provided significant improvements when the speaker adaptation was addressed. Furthermore, results comparable to the current state of the art were reached even when only a limited amount of data was available.

show abstract

Section: Discussionmentioning

confidence: 99%

Speaker-Adapted End-to-End Visual Speech Recognition for Continuous Spanish

Gimeno-Gomez¹,

Hinarejos²

2022

IberSPEECH 2022

View full text Add to dashboard Cite

show abstract

“…Empirical results show a compelling improvement on the performance of eight deep CNN models. We plot the Rank-1 IR of deep [12] 73.3 93.5 98.0 n/a LDMDS [24] 62.7 70.7 65.5 n/a PCLWT [21] 64.76 80.8 74.92 n/a Ghaleb et al [4] n/a n/a n/a 71.7 CNN models for each of six crop ratios as illustrated in Figure 5 for SCFace [5], and Figure 6 for ICB-RW [15] benchmarks. Table 3 summarizes the Rank-1 IR results achieved by 1.30 crop ratios for the eight deep models.…”

Section: Effect Of Increasing the Amount Of Informationmentioning

confidence: 99%

“…They reported 89% Rank-1 IR for distance 3 (1.0m) of SC-Face [5]. A Patch Based Cascaded Local Walsh Transform (PCLWT) followed by whitened principal component analysis is employed in [21] for feature extraction. They reported 64.76%, 80.8%, and 74.92% Rank-1 IR for d1, d2, and d3 respectively.…”

Section: Related Workmentioning

confidence: 99%

Exploring Factors for Improving Low Resolution Face Recognition

Aghdam¹,

Bozorgtabar

Ekenel³

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

State-of-the-art deep face recognition approaches report near perfect performance on popular benchmarks, e.g., Labeled Faces in the Wild. However, their performance deteriorates significantly when they are applied on low quality images, such as those acquired by surveillance cameras. A further challenge for low resolution face recognition for surveillance applications is the matching of recorded low resolution probe face images with high resolution reference images, which could be the case in watchlist scenarios. In this paper, we have addressed these problems and investigated the factors that would contribute to the identification performance of the state-of-the-art deep face recognition models when they are applied to low resolution face recognition under mismatched conditions. We have observed that the following factors affect performance in a positive way: appearance variety and resolution distribution of the training dataset, resolution matching between the gallery and probe images, and the amount of information included in the probe images. By leveraging this information, we have utilized deep face models trained on MS-Celeb-1M and finetuned on VGGFace2 dataset and achieved state-of-the-art accuracies on the SCFace and ICB-RW benchmarks, even without using any training data from the datasets of these benchmarks.

show abstract

“…A dense local method for the representation of images is known as the Local Walsh Transform (LWT) is discussed by Uzun-Per and Gokmen [1]. The LWT is applied for utilizing the well-known Walsh remodel (WT) of every pixel of a photo.…”

Section: Introductionmentioning

confidence: 99%

Person Recognition by Hilbert Pair of Wavelets using Facial Images

Hemalatha

Логашанмугам

2018

IJET

View full text Add to dashboard Cite

In human identification, the face acts as an important tool that carries the identity of each person. The human mind has the ability to re-cognize faces after the first view of a human face. Though there are many types of face detection/recognition system found no method can give the 100% accurate outputs. In this proposed system we are implementing and analyzing a new method that can be used for person recognition system that can produce better output accuracies. In the proposed system of person recognition method, one of the robust wavelet transform methods is used for the extraction of the features from the original images. The wavelet type used is known as the Dual Tree M-Band Wavelet Transform (DTMBWT) method. Using this transform the low and high sub-bands is obtained. These low and high sub-band coefficients are given as the input for the classification purpose. The sub-band obtained from the DTMBWT transform is given as the inputs for the classification purpose. The classification process is done using the K-Nearest Neighbor (KNN) classifier scheme. The system is implemented by using the facial images from the ORL database. By using this dataset images the performance measures of the proposed system is calculated in the form of graphical results such as Receiver Operating Characteristic (ROC), Inverse ROC and Expected Performance Curve (EPC) curves. Results show that proposed DTMBWT based face recognition provides better results than other approaches.

show abstract

Face recognition with Patch-based Local Walsh Transform

Cited by 12 publications

References 42 publications

Speaker-Adapted End-to-End Visual Speech Recognition for Continuous Spanish

Speaker-Adapted End-to-End Visual Speech Recognition for Continuous Spanish

Exploring Factors for Improving Low Resolution Face Recognition

Person Recognition by Hilbert Pair of Wavelets using Facial Images

Contact Info

Product

Resources

About