Since 2020, I am research associate in Osnabrück university. Here we had two nice projects about Wound detection and Face recognition. From 2021-2023, I taught Advanced topics in deep learning and Machine learning course in Osnabrück university. From 2013-2020, I was lecturer at the Payame Noor University and University of Applied Science and Technology, which I had different courses like: Electronic, Image processing, Signal and system,Linear control, Electronic Labs. My master was Electrical engineering and my interest fields of research are deep learning, knowledge distillation, transfer learning, Feature visualization, Machin learning and medical image processing.

LCW: a dataset for children

Some images of LCW dataset

• You can not access to a public child face dataset?
• Your dataset is noisy?
• You need more identities
• Use the (LCW) dataset
• In klix project, we proposed a novel dataset ”Labeled Children in the Wild” (LCW) for facial recognition tasks, that features are more challenging (high variance) for modern face recognition solutions and freely available.
• We used a model that was pre-trained on VGGFace and fine-tune its weights by training on our dataset (Refer to my paper)

Can AI detect the Wound part?

Detecting the wound part by Xception model

• Using deep neural Xception model, the wound part detected.
• The wound part marked by Grad-cam.(Refer to my paper)

Can you realize the real one?

• I guss you could’t realize that. In this figure, the left one is real and the right one is synthetic. We generated such thoes Images.
• In ZIEL project, we generated 100,000 synthetic wound Images using StyleGan3. (Refer to my paper).

Malaria parasite detection

Extracting red blood cell mask, left is the blood cell image, right is red blood cell mask

• A method to detect malaria parasite in blood samples stained with giemsa.
• Most of malaria parasites exist in red blood cells.
• At the first step, the red blood cell mask is extracted.
• Color histogram, granulometry, gradient and flat texture features are extracted
• Different classifires and combination methods applied (Refer to my paper)