Wherein region (mostly based on the moment):
Area, width, height, aspect ratio, minimum bounding rectangle (circle), the centroid, roundness.
Convex hull, convexity (0-1, area / area of the convex hull, for measuring the degree of a compact area. The result is usually highly undesirable non-convex).
Compactness (≧ 1, the contour perimeter ² / (4π area), the circle compactness = 1). Convexity of suitable inflection point, for compactness rounded.
Gradation characteristic (mostly based on statistics):
The average gray value (the overall brightness), the standard deviation, the gradation value of gray value difference (measures how much texture).
Histogram, a cumulative histogram gradation.
Normalized gray value (decrease the light sensitivity)
[Reference] machine vision algorithms and applications, Carsten Steger, Markus Ulrich, Christian Wiedemann with, Yangshao Rong translation, Tsinghua University Press.