AI Bias and Poor Tuning Lead to ‘Unsatisfactory’ Portrait Mode Images for Certain Skin Tones, DxOMark Study Finds
Portrait mode photography, a popular feature on modern smartphones, has been criticized for delivering suboptimal results for subjects with darker skin tones, according to a new study. This feature, designed to enhance images by digitally blurring the background and focusing on the subject, has become a staple in mobile photography. However, the study highlights significant disparities in image quality based on the skin tone of the subject.
The research, conducted by DxOMark, a leading authority on smartphone camera testing, reveals that portrait mode can exhibit biases that affect how well images of individuals with darker skin tones are rendered. The study, which involved flagship smartphones from late 2022 and early 2023, assessed user preferences by analyzing 405 scenes captured by 83 regular consumers. The findings suggest that the portrait mode on many devices produces less satisfactory results for darker skin tones compared to lighter skin tones.
One key issue identified in the study is the AI bias present in the training of portrait mode algorithms. These algorithms are often optimized based on data that may not fully represent the diversity of skin tones. As a result, the portrait mode may struggle with over-exposing the subject’s skin, under-exposing the background, or improperly enhancing skin brightness. This can lead to images that do not accurately reflect the subject’s appearance.
The DxOMark study introduced a ‘Satisfaction Index’ to quantify user experiences with portrait mode. The index revealed that while overall satisfaction with portrait mode was lower than anticipated, the disparity in image quality for darker skin tones was particularly pronounced. This discrepancy underscores the need for more inclusive training data and better tuning of AI algorithms to address these biases.
The implications of these findings are significant for both consumers and manufacturers. For users, particularly those with darker skin tones, the quality of portrait mode images can impact their overall satisfaction with a smartphone’s camera capabilities. For manufacturers, there is a growing need to address these biases in their camera software to ensure a more equitable and accurate representation of all users.
In conclusion, the DxOMark study highlights a crucial area for improvement in smartphone camera technology. As the industry continues to advance, addressing AI bias and improving the tuning of portrait mode features will be essential for achieving fairness and inclusivity in mobile photography.