A recurrence plot (RP) is a straightforward-from-visualize characteristics of similar system states attained at Differen T times (Eckmann et al., 1987).
, that is, the RP recognizes a state pair that is scaled in time. Cross-recurrent Plot, which forms a two-dimensional axis, the black in the coordinate system represents a similar state pair, in fact in this coordinate system, most of the points are not black. The path on the diagonal represents a similar sub-sequence of two sequences.
Obviously, (a) is more similar to the (b) sequence pair. However, this is only the discrimination of the human eye, although the fact is also true, but it must still need quantitative discriminant. For example, to make an image two classification of the classifier, to distinguish between two pairs of graphs which more matching, but the problem is such a sparse black and white two-color map, which features to extract what is suitable for classification? Based on this consideration (in fact, based on a large number of experiments, the characteristics of the classification are not good to extract), the researchers have decided on their own calculation methods, such as DP:
Limiting the slope of the path to 1 is too dead, but if you change it to 0 to infinity and too big, take a compromise strategy:
The path slope of the compromise strategy is [1/2, 2], and the strategy of DTW is similar to that of any, in fact, the problem of application to dynamic programming has this problem and similar solution. In short, now the direction of the path to grasp, and then solve the details of the problem. The problem is that if there is no match between the two consecutive paths because of the very individual points, it is too bad to miss the chance to form the optimal path. Therefore, in order to solve this problem, the attenuation variable is introduced, so that if there are individual points do not match, the path will not be cut directly, but to the path to hit the chapter lower:
The attenuation should not be too small, at least 1 bigger. OK, that would be a good fit.
[Sequence Alignment Methods] Cross-recurrent Plot (CRP)