Bidimensional Distribution Entropy to Analyze the Irregularity of Small-Sized Textures
Hamed Azami, Javier Escudero, Anne Humeau‐Heurtier · IEEE Signal Processing Letters · 2017
Two-dimensional sample entropy (SampEn2D) has been recently proposed to quantify the irregularity of textures. However, when dealing with small-sized textures, SampEn2Dmay lead to either undefined or unreliable values. Moreover, SampEn2Dis too slow for most real-time applications. To alleviate these deficiencies, we introduce bidimensional distribution entropy (DistrEn2D). We evaluate DistrEn2Don both synthetic and real texture datasets. The results indicate that DistrEn2Dcan detect different amounts of white Gaussian and salt and pepper noise, and discriminate periodic from synthesized textures. The results also show that DistrEn2Ddistinguishes different kinds of textured surfaces. In addition, DistrEn2D, unlike SampEn2D, does not lead to undefined values. Moreover, DistrEn2Dis noticeably faster than SampEn2D. Overall, DistrEn2D-as an insensitive feature extraction method to rotation-is expected to be very useful for the analysis of real image textures.