Table 1 Description for the features used in the proposed SRC framework

Texture

Local binary pattern (LBP) [23–25]

Uniform LBP histograms are computed from the segmented object; LBP operator with a circularly symmetric neighbourhood of P members on a circle radius of R is employed; the three-resolution combination is used by setting LBP parameters (P,R) values of (8,1), (8,2), and (8,3)

354

Spatial gray level dependence (SGLD) [26]

13 features, namely, "correlation", "energy", "entropy", "inertia", "inverse difference moment", "sum average", "sum variance", "sum entropy", "difference energy", "difference variance", "difference entropy", "information measure of correlation 1", "information measure of correlation 2" are extracted from each SGLD matrix at six different inter-pixel distances (d = 1, 2, 4, 6, 8, and 10) and in four directions ($\theta =0^{\circ },45^{\circ },90^{\circ },$and $135^{\circ }$), are used to calculate 24 SGLD matrices, yielding 312 SGLD features

312

Run length statistics (RLS) [27]

Five features, namely, "short run emphasis", "long runs emphasis", "gray-level nonuniformity", "run-length nonuniformity", and "run percentage" are obtained from the gray level run length matrices with four directions, $\theta =\left\{0^{\circ },45^{\circ },90^{\circ },135^{\circ }\right\}$

20

Gray level difference statistics (GLDS) [28]

Four features "contrast", "angular second moment", "entropy", and "mean" are extracted from the gray level difference statistics vector; six different inter-pixel distances (d = 1, 2, 4, 6, 8, and 10) and four directions ($\theta =0^{\circ },45^{\circ },90^{\circ },$and $135^{\circ }$) are used to calculate 24 GLDS vectors, yielding 96 GLDS features

96

Shape

Normalized radial length (NRL) [29]

NRL mean, NRL standard deviation, NRL area ratio, NRL zero crossing count, NRL entropy

5

Intensity

[11]

Contrast measure, Average gray level, Standard deviation, Skewness, Kurtosis

5

Spiculation

Region-based stellate features [30]

Means of pixel-wise stellate features are computed from the three local regions (core, inner, and outer regions, respectively); standard deviation of means of pixel-wise stellate features are computed from the three local regions; differences of means of pixel-wise stellate features are computed from the three local regions

20