Represent Contrast using Multidimension Histogram?
Does spatial contrast can be measured by histogram analysis? I am thinking about this problem because histogram represents the global representation. My objective is to detect object-based visual attention with scale invariance that means to analyze the attention of two objects with different size. From the human vision knowledge, different granularity (window size) should be used for different object.
First thought about this problem is something similar to the famous co-ocurrance matrix. In my view, it can be used to estimate the scale of the current location. I argue that the value of several main diagonals of a series of co-ocurrance matries indicates the scale of centered object.
First thought about this problem is something similar to the famous co-ocurrance matrix. In my view, it can be used to estimate the scale of the current location. I argue that the value of several main diagonals of a series of co-ocurrance matries indicates the scale of centered object.
Draft Idea: Suppose the location P(x, y) is measured currently
(1). Calculate a series of co-ocurrance matries in the local neighborhood of P;
(2). Check the three (-1, 0, 1) main diagonals of each matrix. When the elements on the main diagonals become small, the corresponding displace vector is the average scale radius of the scale
