Skin troubles and disorders such as acne, pigmentation, and eczema not only affect appearance but also cause physical discomfort and influence self-perception. These challenges have driven significant research in dermatology, particularly in the development of treatment strategies and skincare products. However, the simulation and prediction of their progression remain relatively underexplored. Improving the ability to model the progression of skin disorders can support clinical decision-making, enhance patient awareness, and promote treatment adherence, thereby contributing to progress in dermatological research. In this study, we propose a pipeline for simulating the progression of skin disorders using generative models. Our approach involves fine-tuning a diffusion model to create a comprehensive multi-paired dataset of skin disorders and training a GAN to simulate severity transitions. As a result, we generate photorealistic images that accurately depict the progression of skin disorders with precise control over severity levels.

Facial acne is a common skin condition that can easily occur in oily skin. Since acne is a small area and its occurrence largely depends on the skin condition, it is difficult to detect it accurately. In this paper, we propose a new method for detecting facial acne based on semantic segmentation. As the layers in a typical CNN-based deep learning model become deeper, the spatial dimension decreases and the number of channels increases. However, since acne is a small object, its spatial information can be lost as the layers get deeper, and it is critical for detecting facial acne. To alleviate this problem, we propose a center point loss, which maintains the center of the acne even in the reduced spatial dimension of the layers and improves the detection performance. First, we generated a center point ground truth indicating the center of each acne from an acne ground truth and applied two max-pooling layers having different kernel sizes, respectively. Following that, center point maps were obtained from the acne segmentation model's two deep decoders, and center point losses were calculated with the center point ground truth. In addition, a semantic segmentation loss was computed by comparing the final feature map and the acne ground truth. Finally, we used the center point losses and the segmentation loss to train the acne segmentation model. Our proposed method was tested using images obtained from a commercial image acquisition system using facial skin analysis equipment. In our experiments, the performance was improved by using the center point loss, which showed higher IoU performance than the existing deep supervision method that used multi-losses for decoders.

Facial wrinkles are important indicators of human aging. Recently, a method using deep learning and a semi-automatic labeling was proposed to segment facial wrinkles, which showed much better performance than conventional image-processing-based methods. However, the difficulty of wrinkle segmentation remains challenging due to the thinness of wrinkles and their small proportion in the entire image. Therefore, performance improvement in wrinkle segmentation is still necessary. To address this issue, we propose a novel loss function that takes into account the thickness of wrinkles based on the semi-automatic labeling approach. First, considering the different spatial dimensions of the decoder in the U-Net architecture, we generated weighted wrinkle maps from ground truth. These weighted wrinkle maps were used to calculate the training losses more accurately than the existing deep supervision approach. This new loss computation approach is defined as weighted deep supervision in our study. The proposed method was evaluated using an image dataset obtained from a professional skin analysis device and labeled using semi-automatic labeling. In our experiment, the proposed weighted deep supervision showed higher Jaccard Similarity Index (JSI) performance for wrinkle segmentation compared to conventional deep supervision and traditional image processing methods. Additionally, we conducted experiments on the labeling using a semi-automatic labeling approach, which had not been explored in previous research, and compared it with human labeling. The semi-automatic labeling technology showed more consistent wrinkle labels than human-made labels. Furthermore, to assess the scalability of the proposed method to other domains, we applied it to retinal vessel segmentation. The results demonstrated superior performance of the proposed method compared to existing retinal vessel segmentation approaches. In conclusion, the proposed method offers high performance and can be easily applied to various biomedical domains and U-Net-based architectures. Therefore, the proposed approach will be beneficial for various biomedical imaging approaches. To facilitate this, we have made the source code of the proposed method publicly available at: https://github.com/resemin/WeightedDeepSupervision.

Facial wrinkle is very important in measuring aging. Image processing-based methods have been proposed for wrinkle detection, but their performance was not enough because wrinkles have a wide variety of thickness, shape, orientation, and vague boundaries. Recently, deep learning-based methods have been widely applied in the field of image recognition with a lot of labeled image dataset. To extend this technology to facial wrinkle detection, labeling work for wrinkles to generate ground truth is very important. However, it is difficult to label wrinkles accurately because of the wide variety. In this paper, we propose a semiautomatic labeling strategy incorporating a texture map and a deep learning model. Specifically, the proposed method extracted the texture map from an original image and removed non-wrinkle textures on the map by multiplying with a roughly labeled wrinkle mask. Then, the map is converted into ground truth by thresholding. Using the ground truth, a deep learning model was trained with the original image and the texture map. The trained model was evaluated with facial images obtained from real skin diagnosis devices, and the results showed superior performance to those of existing image processing-based methods.

Poresare minute skin openings through which hair and sebum come out and appear as holes in the facial skin. Enlarged pore is one of the major concerns for people who care about their skin. Remedies include the use of cosmetics and pore-reduction medical procedures. Awareness of the condition of one's facial pores and appropriate management are required to prevent pore deterioration. Pore segmentation algorithms based on classical image processing are characterized by low accuracy and high computational costs. In addition, these algorithms require that input images be taken in light-controlled environments. These issues were resolved by using a light-specialized data augmentation method and a neural network with a narrow receptive field for identifying local features. We introduce Pore-Net, an algorithm that can be used on mobile devices to segment pores with a low computational cost, using selfie-camera images as an input. Pore-Net has the following algorithm flow. First, a confidence map-based segmentation without encoder-decoder form is applied to lower the computational costs on high-resolution input images. Second, pre- and post-processing for input based on region-of-interest(ROI) of facial landmarks are performed to work robustly in mobile devices. Pore-Net achieved the lowest computational cost in inference time and multiply-and-accumulates(MACs) when compared with the binary segmentation models with similar performance in intersection-over-union(IoU).

Techniques for video fingerprinting are helpful in managing vast libraries of video clips. Recent advances have shown that video tomography and Bag-of-Visual-Words (BoVW) can be successfully used for the purpose of video fingerprinting. In this paper, we introduce a novel video signature (i.e., a novel video fingerprint) that takes advantage of both video tomography and BoVW. Specifically, the proposed video signature is created by first extracting inclined tomography images from the video content, and by subsequently applying the BoVW approach to the inclined tomography images obtained. The key to our approach is that we make the angle of inclination of the tomography images dependent on the amount of motion in the video content. That way, the proposed video signature is able to capture both spatial and temporal information. Experimental results obtained for the publicly available TREVID-2009 video set indicate that video copy detection by means of the proposed video signature is robust against spatial and temporal transformations.