1: \begin{abstract}

2: 

3: We investigate the performance of entropy estimation methods, 

4: based either on block entropies or compression approaches, 

5: in the case of bidimensional sequences.

6: We introduce a  validation dataset made of %black and white

7: images produced by a large number of different 

8: natural systems, in the vast majority characterized by long-range correlations,

9: which produce a large spectrum of entropies.

10: Results show that the framework based on lossless

11: compressors applied to the one-dimensional projection 

12: of the considered dataset leads to poor estimates. 

13: This is because higher-dimensional 

14: correlations are lost in the projection operation.

15: The adoption of compression methods which do not introduce

16: dimensionality reduction improves the 

17: performance of this approach.

18: %{\bf Anyway, the traditional block-entropies method

19: %shows a really faster convergence 

20: %and evident better results in estimating 

21: %the asymptotic  entropy.}

22: %{\bf 

23: By far, the best estimation of the asymptotic  entropy is

24: generated by the faster convergence of the traditional block-entropies method.

25: %}

26: As a by-product of our analysis, we show how a specific

27: compressor method can be used as a potentially interesting  %approach 

28: technique

29: %algorithm 

30: %to be applied to the difficult problem of 

31: for automatic detection of symmetries in textures and images.

32: \end{abstract}

33: