Robust Extraction of Human Facial Components Using a Landmark-based Model
Main Article Content
Background/Objectives: Unlike general cameras, high-speed cameras capable of capturing a very large number of frames per second can enable the advancement of image processing technologies that have been limited.
Methods/Statistical analysis: In this paper, we propose a method of removing noises from high-speed color images and then detecting human facial areas from the noise-removed image. In this paper, first, noise pixels included in the ultrafast image are effectively removed by applying a bidirectional filter. Then, using a retina face model, a face region representing a person’s personal information is robustly detected from an image from which noise has been removed.
Findings: Experimental results show that the proposed algorithm removes noises from images and then robustly detects the face region using the generated model. In this study, the performance of the proposed model-based face detection approach was quantitatively compared and evaluated in terms of accuracy. In this study, the accuracy scale expressed as the ratio of the number of face regions accurately extracted through the introduced method and the number of face regions originally existing in the entire image data was used. For performance evaluation, we also implemented the method using the existing fixed model. The existing face detection has not been able to compensate for noise included in images. In addition, since an inflexible model was used, many errors occurred in face detection. In contrast, the proposed method removes undesirable noise contained in an image by applying a bidirectional filter. Then, a flexible model composed of five landmarks is created to detect a face from an image from which noise has been removed, so that accurate results can be obtained.
Improvements/Applications: The proposed face detection method is expected to be used for many application fields related to pattern recognition such as building monitoring, door management, and mobile biometric authentication.
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.