Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session E

Poster F1 - Standardizing Color Accuracy for Smartphone Videos with Capture Angle

Saturday, October 26, 2024

10:00 AM - 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

RH

Ruijing Han

Undergraduate Research Assistant
Vanderbilt University
Brentwood, Tennessee, United States

Co-Author(s)

IS

Inga Saknite, PhD (she/her/hers)

Adjoint Assistant Professor
Vanderbilt Dermatology Translational Research Clinic, United States
ET

Eric Tkaczyk

Assistant Professor
Vanderbilt Dermatology Translational Research Clinic, United States

Primary Investigator(s)

Audrey K. Bowden, PhD

Professor
Vanderbilt University, United States

Introduction: With the growing popularity of telehealth, smartphone cameras have expanded the options for more portable diagnostic opportunities for clinicians, especially in dermatology. However, the coloring of the skin plays an important role in assessing for skin irregularities that indicate a potential dermatological disease or other health conditions. Without proper skin color assessment, it could lead to missed affected skin areas or underdiagnosis, which can impact the patient’s treatment. As a result, accurate skin color representation in smartphone photography is vital for the proper assessment of dermatological conditions. Multiple studies have shown that factors such as camera device, lighting, distance, and angle can affect the perceived color of the skin. Despite these findings, there is not a clear method that allows for the identification and adjustment of the ‘actual’ underlying color across all these parameters. To address this gap, we aim to develop a color calibration method that corrects the varying color appearances to a standardized color, accounting for distance and angle with videos.

Materials and

Methods: Skintone swatches (n = 21) from the Pantone Skintone Guide were assessed, with calibration swatches (n = 15) serving as the color standard and validation swatches (n = 6) used for verification.
To measure the color effect from angle changes, calibration and validation swatches were placed on a white card as a backdrop and positioned vertically on a rotation stage. A vertically mounted smartphone (iPhone SE 2nd Gen) was placed 36 cm from the swatches. Videos of swatches rotated from -45 to 45 degrees from the camera were taken with the phone’s flash as the sole light source in the room. For video processing, MATLAB 2024a was used to identify and track the swatches throughout each video frame to measure the average color in CIEL*A*B color space, where values for L*, A*, and B* channels were collected. These values were considered the pre-processed data.
The difference between the ‘ground-truth’ A*/B* values and the collected A*/B* values for each angle was calculated for each Pantone calibration swatch to determine the required correction value. A spline curve fit was applied to these correction values for each calibration swatch. A 3D model was created to fit the correction spline curves for all 15 swatches, predicting the needed correction value using angle and pre-processed A*/B* values. For affirmation, the model was used to correct the collected A*/B* values of the validation swatches based on their respective angles, which were considered post-calibrated values. A paired t-test was conducted to evaluate the success of the method.

Results, Conclusions, and Discussions: The average between all the validation Pantones of the absolute value of the difference between the initial A* value extracted from the video and the ground-truth A* value as well as between the post-calibrated A* value and the ground-truth A* value were found to be 4.8367 +/- 1.0638 and 1.5164 +/- 0.7784, respectively. The same was done for the respective B* values, with a pre-processed value difference of 6.3787 +/- 0.9578 and a post-calibrated value difference of 1.5119 +/- 1.1635. For both A* and B* values, there is a decrease in difference after the color values are adjusted with the model. Further analysis with paired sample t-test showed a significant difference between the video-collected data and post-calibrated difference for both A* (t = 4.8351, p=0.0047, df = 5) and B* values (t = 9.4804, p=0.0002, df = 5). However, the difference from the ground-truth value for both prior and after correction varies for the individual swatches, as given by the standard error measurements (SEM) of 0.9578 and 1.1635 after correction for the A* value and B* value, respectively.
The p-values of less than 0.05 with the post-calibrated difference lower than that collected from the video indicate the method can decrease the distance from the ground-truth value for the validation swatches. Hence, the method provides a way to determine a more accurate reconstructed color value of the Pantone swatches than that directly extracted from the smartphone camera. With the required correction value differing slightly between angles, applying color correction based on a single or unknown angle does not provide a full-color reconstruction of an examined area and could introduce bias, skewing the corrected color value. Although there is not a near zero average of the difference from the corrected data to its corresponding ground truth value, using multiple angles is a step closer to more accurately predicting the color values from video data on account of the changes in the viewing angle of the examined object. Further investigation into this method could be conducted by extracting color data from RAW video, which avoids the innate camera processing typically applied after recording.

Acknowledgements (Optional):