Fig. 1
From: Improved gradient local ternary patterns for facial expression recognition
a Source image. b Horizontal and vertical Sobel masks. c Horizontal and vertical gradients from Sobel operator. d Gradient magnitude image. e Sample 3×3 neighbourhood of gradient magnitudes. f GLTP codes for t=10. g Splitting GLTP code into positive and negative code. h Positive and negative decimal coded image. i Segmenting positive and negative images into 7×6 regions. j Concatenate positive and negative GLTP histogram from each region together to form feature vector