Login | Signup

Comprehending Developmental Prosopagnosia Using a Trained Deep Convolutional Neural Network: A Perceptual Study

Eushra Rahman

July 25, 2025

ISBN: 979-8-89480-841-3


Developmental prosopagnosia (DP), also known as face blindness, is a condition in which individuals have an impairment in the FFA region of their brain. This results in inability to recognize faces. Majority of the current body of literature focuses on the biological aspect of developmental prosopagnosia, but there is a scarcity of articles relating the condition to a perceptual standpoint. The goal of this study is to see how images depicting the visual impairments seen by people with developmental prosopagnosia affect the robustness of a convolutional neural network trained to exhibit normal vision. This was done through the use of a qualitative coding scheme to extract 2 criteria: items needed for recognition and perception of faces amongst ten individuals with DP. The results of the coding scheme demonstrated that all of the individuals used external cues such as environment, age, stylistic choice , hair, etc. to identify others. The second coding scheme was used to extract commonalities between how individuals saw the faces. This scheme showed three main themes: blurred faces, non-distinguishable facial features, and distinguishable facial features. Both of these qualitative results were used to create 3 visualizations of what individuals with the condition perceive based on the common themes examined between the ten individuals. These renditions were passed through Google AutoML’s convolutional neural network VertexAI, which was trained using the CelebA dataset (public domain). The results demonstrated how 2 out of the three renditions negatively affected the network’s ability to identify the images as faces (see Fig. 8 and Fig. 9). This study provides an overview of the condition from a perceptual point of view, which is vital in order to educate others about the implications of the condition.

References