Comparison of colour transforms used in lip segmentation algorithms

摘要: Lip segmentation is a fundamental system component in range of applications including: automatic lip reading, emotion recognition and biometric speaker identification. The first step involves applying colour transform to enhance the contrast between lips surrounding skin. However, there much debate among researchers as best for this task. As such, article presents most comprehensive study date by evaluating 33 transforms segmentation: 21 channels from seven space models (RGB, HSV, YCbCr, YIQ, CIEXYZ, CIELUV CIELAB) 12 additional (8 which are designed specifically segmentation). comparison extended determine segment oral cavity. Histogram intersection Otsu’s discriminant used quantify compare transforms. Results lip–skin validate experimental approach, 11 top literature. necessity selecting correct demonstrated an increase accuracy up three times. Hue-based including pseudo hue domain filtering perform segmentation, with HSV achieving greatest 93.85 %. a* CIELAB performs lip–oral cavity while LUX reasonably well both segmentation.