Elkarazle, K.; Raman, V.; Then, P. Facial Age Estimation Using
Download 0.59 Mb. Pdf ko'rish
|
BDCC-06-00128
- Bu sahifa navigatsiya:
- 3. Common Challenges
2. Contributions
The contributions to the body of the literature are summarised below: 1. Several architectures and methods used to build facial age classifiers are reviewed. 2. Several common challenges faced by researchers are described. 3. An analysis of the benchmark datasets used to build age estimation systems is presented. 4. A survey on several recently proposed age estimation methods is provided. 5. A discussion of the existing gaps and the trends based on the surveyed papers is demonstrated. 3. Common Challenges There are many challenges researchers encounter when building age estimation mod- els. Some of these challenges are considered “controllable”, in which further tweaking of the model or introducing new measures is required to overcome them. Examples of these controllable challenges include head pose, image quality, or image resolution. In contrast, “uncontrollable” challenges are those that scientists at the current stage of machine learning research cannot control; thus, creating significant roadblocks to achieving better age estimation accuracies. Big Data Cogn. Comput. 2022, 6, 128 3 of 22 3.1. Head-Pose and Alignment One of the common controllable challenges is the head pose and alignment, which refers to the position of the face in a given image. As images are captured in real-life conditions, the position of faces tends to vary in terms of alignment. This issue is usually resolved during the pre-processing stage, in which the face is detected and realigned. In [ 2 ], the authors reported that head poses and alignment contributed to the decrease in performance. 3.2. Image Resolution In any computer vision problem, the resolution of training samples plays a significant role in shaping the accuracy of an age estimation model. Images of lower quality tend to lose critical ageing features such as wrinkles or the shape of a face, preventing the model from learning all the necessary discriminative features to tell ages apart. This issue is caused by the various resolution image capturing devices poses; however, it can be solved by normalising the resolution of both training and testing samples through different image upsampling methods. Researchers in [ 3 ] have shown that enhancing the resolution of the images prior to training improves classification accuracies. 3.3. Lifestyle and Health Condition A person’s lifestyle is one of the considerable uncontrollable challenges that prevent most age estimation models from performing well. Depending on one’s health condition and lifestyle, they can appear older or younger than their actual age; thus, confusing the model that attempts to predict their age. 3.4. Lack of Data Overfitting is a severe issue in age estimation, and it occurs as a result of the lack of diverse facial images for training. Although large-scale facial datasets exist for various facial analysis problems, most samples are either unlabelled or do not cover enough examples, such as subjects from different genders and ethnicities. Several studies such as [ 4 ] and [ 5 ] have stated that the lack of data is the main cause of most of the misclassified samples. 3.5. Genetics The ageing pattern of an individual depends primarily on one’s genes. This issue is one of the roadblocks that existing machine learning models cannot solve since it depends on the person’s background, gender, and the climate in which they grew up. In [ 6 ], the researchers showed a relationship between the subject’s gender and their facial age. 3.6. Facial Modifications The presence of facial accessories or facial hair often causes certain features to become unclear or disappear entirely from an image. This issue is considered uncontrollable; how- ever, there is room for more research to develop a model capable of normalising all images by removing particular facial accessories such as beards, piercings, or makeup. The findings in [ 7 ] demonstrated a relationship between the degraded accuracy and facial modifications. Download 0.59 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling