As demand for remote identification via surveillance grows, survey of academic work on biometrics, including facial recognition and gait recognition, reveals gaps in understanding of processing blurry, poorly lit images , at a difficult angle or where the subject has been partially cut out of the frame or is simply too small. Ways to improve methods and streamline algorithms that will help bring better recognition to the edge are needed.
“An investigation of face- and body-based human recognition that is robust against image blur and low illumination” from the Electronics and Electrical Engineering Division of Seoul Dongguk University, published by MDPI , focuses on global biometric “human recognition” from problematic images and attempts to improve image quality to counter the problems.
While facial recognition has received much of the research attention because faces are believed to contain the most important information for identification, researchers say that multimodal body and gait recognition can help what they call “global human recognition”.
Low resolution images have been dealt with sufficiently, but the blurry image survey articles are not exhaustive, which the article attempts to address. It reviews studies on blurry image restoration and low light and categorizes them by whether or not deep learning was used and whether face and body were combined.
The team tackles indoor and outdoor environments that generate distinct problems. Indoor images are more prone to motion blur and difficult angles when subjects are closer. Outdoors, lighting may be uneven and lower resolution images.
“No studies have yet been conducted on body-based human recognition robust to image blurring in indoor environments; in this case, only the body region is used, dismissing the face. In other words, neither body-based recognition nor body-based re-identification has yet been investigated,” the study states. “This is because, compared to the face region, the body region requires more global characteristics for recognition. This in turn implies that recognition performance is not significantly affected by image blur.
The survey covers how the degree of blurriness is assessed through image quality assessment, how clothing color can affect body recognition results. Gait recognition can be used in more situations due to the low impact of image blur.
Further studies are needed on the impact of low lighting on gait-based recognition. Further studies are also needed for human recognition in more severe low-light cases. So far, studies have avoided this because “it is difficult to restore color perfectly when converting severe low-light images to normal-light images. It is expected that these problems can be solved through the different deep learning methods.
“Overall, facial and body recognition show higher accuracy and longer processing time than the face-based method,” say the authors, who hope to help biometrics researchers as more and more applications are made for the detection of criminals and missing persons. , as well as parameters such as driver identification in vehicles and crowd analysis.
biometric identification | biometrics | biometric research | facial recognition | walking recognition | multimodal | re-identification | video surveillance