First Order Fuzzy Transform for Images Compression

In this paper, we present a new image compression method based on the direct and inverse F1-transform, a generalization of the concept of fuzzy transform. Under weak compression rates, this method improves the quality of the images with respect to the classical method based on the fuzzy transform.

• Fuzzy Transform
• Generalized Fuzzy Partition
• Basic Function
• Hilbert Space
• Image Compression
• PSNR

1. Introduction

We present a new image compression method based on the discrete direct and inverse F¹-transform which is a generalization of the classical fuzzy transform [1] [2] identified as F⁰-transform (for brevity, F-transform).

The F-transform compression technique [3] is a lossy compression method used in image and video analysis [4] - [18] and in data analysis [19] - [25] as well. In [26] , the concept of the F-transform was extended to the cases with various types of fuzzy partitions. In [1] [27] , the F^s-transform (s ≥ 1), a generalization of the F-transform, was presented: in other terms, the constant components of the F-transform were replaced by polynomials in order to capture more information of the original function. In particular, the F¹-transform was used for the edge detection problem [1] [2] . The aim of this paper is to improve the quality of the decoded images after their compression via the F¹-transform-based method.

Strictly speaking, we divide images of sizes N × M into smaller images (called blocks) of sizes N(B) × M(B) and then we code each block into another one of sizes n(B) × m(B), where n(B) < N(B) and m(B) < M(B). The compression is performed by calculating the direct F¹-transform components with first degree polynomials. Afterwards, we calculate the inverse F¹-transform and obtain the corresponding decoded blocks, recomposed to obtain the final reconstructed image. In Figure 1, we describe this process in detail.

The compression rate is given by. The quality of a decoded image is measured by the Peak Signal to Noise Ratio (PSNR) index.

In Section 2, we recall the definition of h-uniform generalized fuzzy partition and the concept of F¹-transform. In Section 3, a F¹-transform-based compression method is presented and it is applied to images considered as fuzzy relations: there every image is partitioned into smaller blocks and the direct and inverse F¹-transforms are calculated for each block. Then the decoded blocks are recomposed and the PSNR index is calculated. In Section 4, tests are applied to grey image datasets and the results are compared with similar results obtained by using the classical F-transform compression method. Section 5 contains the conclusions.

2. Generalized Fuzzy Partition and F 1 -Transform

We recall the main concepts [2] that will be used in the sequel. We consider a set of points (called nodes) of such that . We say that the fuzzy sets form a generalized fuzzy partition of, if for each, there exist such that, and the following constraints hold:

Figure 1

The F 1 -transform image compression method

1) (locality) if and if ,

2) (continuity) A_k is continuous in,

3) (covering) for each,.

The fuzzy sets are called basic functions. If the nodes are equidistant, i.e. for, where, if and the following additional properties hold:

4), and for each and,

5) and for every and, then is called an -uniform generalized fuzzy partition. In this case we can find a function, called generating function, which is assumed to be even, continuous and positive everywhere except on the boundaries, where it vanishes, in such a way we have that for:

If, then the -uniform generalized fuzzy partition is said h-uniform generalized fuzzy partition. We can extend the notion of h-uniform generalized fuzzy partition from an interval to the rectangle, so that we have the family of basic functions, where is the product of the corresponding functions from the h₁-uniform generalized fuzzy partition of and from the h₂-uniform generalized fuzzy partition of. Then we can say that is an h-uniform generalized fuzzy partition of, where. In the sequel we consider only such h-uniform generalized fuzzy partitions.

Let be a basic function of and be the Hilbert space of square integrable functions (reals) with weighted inner product:

Likewise, we define the Hilbert space of square integrable in two variables functions with weighted inner product:

Two function are orthogonal if. Let and, be two linear subspaces of and with orthogonal basis given by polynomials and, respectively.

We consider an integer and all pairs of integers (i, j) such that. We introduce a linear subspace of having as orthogonal basis the following:

where s is the maximum degree of polynomials. For s = 1, the orthogonal basis of the linear space is the following:

Let be a set of functions such that for, , , where the function is the restriction of f on . Then the following theorem holds:

Theorem 1. ( [2] , lemma 5). Let. Then the orthogonal projection of on, , is the polynomial of degree s given by

for every, where the coefficients are given by

Following [2] , let be an h-uniform generalized fuzzy partition of and. For s = 1, the orthogonal basis of the linear subspace is given by the polynomials:

Let be the orthogonal projection of on given point wise as

for every, where the three coefficients are defined by Theorem 1:

Then the matrix, defined from (8), is called F¹-transform of the function with respect to the h-uniform generalized fuzzy partition. We define the inverse F¹-transform of the function to be a function as

For sake of completeness, we point out the utility of the concept of inverse F¹-transform which stands in the approximation of the function under certain suitable assumptions. For example, we have the following result:

Theorem 2. ( [2] , theorem 14). Let be an h-uniform generalized fuzzy partition of and be the inverse F¹-transform of with respect to this fuzzy partition. Moreover let be four times continuously differentiable on and A_k (resp., B_l) be four times continuously differentiable on (resp.,). Then the following holds for every:

In other words, the Equality (13) says that we can approximate a function in two variables, four times continuously differentiable on, with the inverse F¹-transform (12) unless to O (h²).

3. F 1 -Transform Image Compression Method

We are interested to the case discrete, i.e. we consider functions in two variables which assume a finite number of values in like finite fuzzy relations. Indeed, let R be a grey image of sizes, , being the normalized value of the pixel, that is if N_lev is the length of the grey scale. Let and be two h-uniform generalized fuzzy partitions of and, respectively, where, , , , ,. Slightly modifying (8), then we can define the (discrete) F¹-transform of R the matrix whose entries are defined as

where, , are given as (by rewriting the Equations (9), (10), (11) in the following form, slightly modified):

The Formula (14) is considered as a compressed image of the original image R. can be decoded by using the following inverse (discrete) F¹-transform defined for every as

We divide the image R of sizes in sub-matrices R^B of sizes, called blocks ( [26] [28] ), each compressed to a block of sizes

, , , via the discrete F¹-transform, as Formula (14), of components given by

We rewrite (15), (16), (17) as

The basic functions and form an h-uniform generalized uniform fuzzy partition of and, respectively. They are generated by the basic functions and, respectively. Then we have that

where, , and

where, , ,. In Figure 2, we show the basic functions (23) for N = 16 and n = 4.

The compressed block is decoded to a block

of sizes by using the inverse F¹-transform defined for every as

which approximates the original block R^B. Making the union of all the decoded blocks R^1B, we obtain a fuzzy relation (denoted with) R¹ of sizes. Then we measure the RMSE (Root Mean Square Error) given by

Figure 2

Cosine basic functions (N = 16, n = 4

which implies that PSNR is the following:

4. Test Results

We compare our method with the classical F-transform compression method, but here no comparison is made with the one inspired to the Canny method used in [2] .

For our tests we have considered the CVG-UGR image database extracting grey images of sizes 256 × 256 (cfr., http://decsai.ugr.es/cvg/dbimagenes/). For brevity, we only give the results for three images as Lena, Einstein and Leopard whose sources are given in Figures 3(a)-(c), respectively.

In Table 1, we show the PSNR of the F-transform and F¹-transform methods for some values of the compression rate in the image Lena.

We make the following remarks on Table 1:

− for weak compression rates the quality of the decoded image under the F¹-transform method is better than the one obtained with the F-transform method;

− for strong compression rates the quality of the images decoded in the two methods is similar;

− the difference between the two PSNR’s in the two methods overcomes 0.1 for ρ > 0.25.

In Figure 4, we show the trend of the PSNR for the two methods.

In Figures 5(a)-(d) (resp., Figures 6(a)-(d)), we show the decoded images of Lena obtained by using the F-transform (resp., F¹-transform) for ρ = 0.0.0625, 0.16, 0.284444 and 0.444444, respectively.

Figure 3

(a) Lena; (b) Einstein; (c) Leopard.

(b) (c)Figure 4

PSNR trend for the source image Lena

Table 1

PSNR of the F-transform and F 1 -transform methods for some values of the compression rate in the image Lena

ρ	PSNR F-transform	PSNR F1-transform	(PSNR F1-tr) - (PSNR F-tr)
0.015625	21.088	21.071	−0.017
0.035156	23.558	23.541	−0.018
0.062500	24.551	24.544	−0.007
0.097656	25.791	25.796	0.005
0.140625	26.812	26.823	0.011
0.160000	26.912	26.941	0.029
0.250000	28.431	28.497	0.066
0.284444	29.012	29.125	0.113
0.297521	29.089	29.247	0.158
0.308642	29.108	29.339	0.231
0.390625	29.899	30.141	0.242
0.444444	30.800	31.023	0.223
0.562500	31.121	31.375	0.254

Figure 5

(a) F-tr under ρ = 0.0.0625; (b) F-tr under ρ = 0.16; (c) F-tr decoded (ρ = 0.284444); (d) F-tr decoded (ρ = 0.444444).

(b) (c) (d)Figure 6

(a) F 1 -tr decoded (ρ = 0.0.0625); (b) F 1 -tr decoded (ρ = 0.16); (c) F 1 -tr decoded (ρ = 0.284444); (d) F 1 -tr decoded (ρ = 0.444444).

(b) (c) (d)

In Table 2 and Figure 7, we show the PSNR obtained using the F-transform and F¹-transform methods for some values of the compression rate in the image Einstein: this table confirms the same results obtained for the image Lena in Table 1.

In Figures 8(a)-(d) (resp., Figures 9(a)-(d)) we show the decoded images of Einstein obtained using the F-transform (resp., F¹-transform) method for ρ = 0.0.0625, 0.16, 0.284444 and 0.444444, respectively.

In Table 3 we show the PSNR values obtained using the F-transform and F¹-transform methods for some values of the compression rate in the image Leopard.

Figure 7

PSNR trend for the source image Einstein

Table 2

PSNR results obtained for the source image Einstein

ρ	PSNR F-transform	PSNR F1-transform	(PSNR F1-tr) - (PSNR F-tr)
0.015625	22.2701	22.2679	−0.0022
0.035156	23.4968	23.4952	−0.0016
0.062500	24.3781	24.3764	−0.0017
0.097656	25.6269	25.6265	−0.0004
0.140625	26.9260	26.9320	0.0006
0.160000	28.0048	28.0186	0.0138
0.250000	29.3003	29.4154	0.1151
0.284444	30.0018	30.1252	0.1234
0.297521	30.4054	30.5377	0.1323
0.308642	30.5415	30.7242	0.1827
0.390625	31.0126	31.1888	0.1762
0.444444	32.3841	32.6976	0.3135
0.562500	33.2661	33.5678	0.3017

Figure 8

(a) F-tr decoded (ρ = 0.0.0625); (b) F-tr decoded (ρ = 0.16); (c) F-tr decoded (ρ = 0.284444); (d) F-tr decoded (ρ = 0.444444).

(b) (c) (d)Figure 9

(a) F 1 -tr decoded (ρ = 0.0.0625); (b) F 1 -tr decoded (ρ = 0.16); (c) F 1 -tr decoded (ρ = 0.284444); (d) F 1 -tr decoded (ρ = 0.444444).

(b) (c) (d)

Table 3 confirms the results obtained for the images Lena and Einstein: the quality of the decoded image obtained by using the F¹-transform is better than the one obtained using the F-transform for weak compression rates. In Figure 10, we show the trend of the PSNR index obtained by using the two methods.

In Figures 11(a)-(d) (resp., Figures 12(a)-(d)), we show the decoded images of Leopard obtained by using the F-transform (resp., F¹-transform) method for ρ = 0.0.0625, 0.16, 0.284444, 0.444444, respectively.

In Figure 13, we show the trend of the difference of PSNR by varying the compression rate for all the images in the dataset above considered.

Figure 10

PSNR trend for the source image Leopard

Table 3

PSNR results obtained for the source image Leopard

r	PSNR F-transform	PSNR F1-transform	(PSNR F1-tr) - (PSNR F-tr)
0.015625	17.2997	17.3183	0.0186
0.035156	18.6483	18.6726	0.0243
0.062500	19.6883	19.7067	0.0184
0.097656	20.0131	20.0375	0.0244
0.140625	22.4336	22.4470	0.0134
0.160000	22.9203	22.9892	0.0689
0.250000	24.4041	24.5474	0.1433
0.284444	25.0750	25.2096	0.1346
0.297521	25.2229	25.3673	0.1444
0.308642	25.4181	25.6597	0.2416
0.390625	26.1470	26.3948	0.2478
0.444444	26.6971	26.9762	0.2791
0.562500	27.7235	28.0978	0.3743

Figure 11

(a) F-tr decoded (ρ = 0.0.0625); (b) F-tr decoded (ρ = 0.16); (c) F-tr decoded (ρ=0.284444); (d) F-tr decoded (ρ=0.444444).

(b) (a) (b)Figure 12

(a) F 1 -tr decoded (ρ = 0.0.0625); (b) F 1 -tr decoded (ρ = 0.16); (c) F 1 -tr decoded (ρ = 0.284444); (d) F 1 -tr decoded (ρ = 0.444444).

(b) (a) (b)Figure 13

PSNR trend for all the images in the dataset considered

Summarizing, we can say that the presence of the coefficients of the F¹-transform is negated by noise introduced during the strong compressions, while this effect increases considerably using weak compressions rates.

5. Conclusion

We give an image compression method based on the direct and inverse F¹-transform. The results show that the PSNR of the reconstructed images with the F¹-transform-based compression method is better than the one obtained with the F-transform-based compression. In the tested dataset of images, we find that the difference between the two corresponding PSNR values is greater than 0.1 (resp., 0.25) for ρ = 0.25 (resp., ρ ≈ 0.5). In the next papers, we shall use the F¹-transform in data analysis problems.

Acknowledgements

We also accomplish this research under the auspices of the INDAM-GCNS, Italy. The last author acknowledges a partial support from the European Regional Development Fund in the IT4Innovations Centre of Excellence project (CZ.1.05/ 1.1.00/02.0070).

References

1. Perfilieva, I. and Hodáková, P. (2013) F1-Transform of Functions of Two Variables. Proceedings of EUSFLAT 2013, Advances in Intelligent Systems Research, Milano, 32, 547-553.

2. Perfilieva, I., Hodáková, P. and Hurtik, P. (2014) Differentiation by the F-Transform and Application for Edge Detection. Fuzzy Sets and Systems, 288, 96-114. https://doi.org/10.1016/j.fss.2014.12.013

3. Perfilieva, I. (2006) Fuzzy Transform: Theory and Application. Fuzzy Sets and Systems, 157, 993-1023. https://doi.org/10.1016/j.fss.2005.11.012

4. Di Martino, F., Loia, V. and Sessa, S. (2003) A Method in the Compression/Decompression of Images Using Fuzzy Equations and Fuzzy Similarities. Proceedings of the 10th IFSA World Congress, Istanbul, 30 June-2 July 2003, 524-527.

5. Di Martino, F., Loia, V., Perfilieva, I. and Sessa, S. (2008) An Image Coding/Decoding Method Based on Direct and Inverse Fuzzy Transforms. International Journal of Approximate Systems, 48, 110-131. https://doi.org/10.1016/j.ijar.2007.06.008

6. Di Martino, F. and Sessa, S. (2007) Compression and Decompression of Images with Discrete Fuzzy Transforms. Information Sciences, 177, 2349-2362. https://doi.org/10.1016/j.ins.2006.12.027

7. Di Martino, F., Loia, V. and Sessa, S. (2010) A Segmentation Method for Images Compressed by Fuzzy Transforms. Fuzzy Sets and Systems, 161, 56-74. https://doi.org/10.1016/j.fss.2009.08.002

8. Di Martino, F., Loia, V. and Sessa, S. (2010) Fuzzy Transforms for Compression and Decompression of Colour Videos. Information Sciences, 180, 3914-3931. https://doi.org/10.1016/j.ins.2010.06.030

9. Di Martino, F., Hurlik, P., Perfilieva, I. and Sessa, S. (2014) A Color Image Reduction Based on Fuzzy Transforms. Information Sciences, 266, 101-111. https://doi.org/10.1016/j.ins.2014.01.014

10. Di Martino, F. and Sessa, S. (2013) Coding B-Frames on Color Videos with Fuzzy Transforms. Advances on Fuzzy Systems, 2013, Article ID: 652429. https://doi.org/10.1155/2013/652429

11. Di Martino, F., Loia, V. and Sessa, S. (2012) Fragile Watermarking Tamper Detection with Images Compressed by Fuzzy Transform. Information Sciences, 195, 62-90. https://doi.org/10.1016/j.ins.2012.01.014

12. Di Martino, F., Loia, V. and Sessa, S. (2013) Image Matching by Using Fuzzy Transform. Advances on Fuzzy Systems, 2013, Article ID: 760705, 10 p.

13. Loia, V. and Sessa, S. (2005) Fuzzy Relation Equations for Coding/Decoding Processes of Images and Videos. Information Sciences, 171, 145-172. https://doi.org/10.1016/j.ins.2004.04.003

14. Nobuhara, H., Pedrycz, W. and Hirota, K. (2000) Fast Solving Method of Fuzzy Relational Equation and Its Application to Lossy Image Compression. IEEE Transactions on Fuzzy Systems, 8, 325-334. https://doi.org/10.1109/91.855920

15. Nobuhara, H., Pedrycz, W. and Hirota, K. (2005) Relational Image Compression: Optimizations through the Design of Fuzzy Coders and YUV Color Space. Soft Computing, 9, 471-479. https://doi.org/10.1007/s00500-004-0366-7

16. Nobuhara, H., Hirota, K., Di Martino, F., Pedrycz, W. and Sessa, S. (2005) Fuzzy Relation Equations for Compression/Decompression Processes of Colour Images in the RGB and YUV Colour Spaces. Fuzzy Optimization and Decision Making, 4, 235-246. https://doi.org/10.1007/s10700-005-1892-1

17. Perfilieva, I. and De Baets, B. (2010) Fuzzy Transforms of Monotone Functions with Application to Image Compression. Information Sciences, 180, 3304-3315. https://doi.org/10.1016/j.ins.2010.04.029

18. Vajgl, M., Perfilieva, I. and Hodáková, P. (2012) Advanced F-Transform Based Image Fusion. Advances in Fuzzy Systems, 2012, Article ID: 125086, 9 p. https://doi.org/10.1155/2012/125086

19. Di Martino, F., Loia, V. and Sessa, S. (2010) Fuzzy Transforms Method and Attribute Dependency in Data Analysis. Information Sciences, 180, 493-505. https://doi.org/10.1016/j.ins.2009.10.012

20. Di Martino, F., Loia, V. and Sessa, S. (2011) Fuzzy Transforms Method in Prediction Data Analysis. Fuzzy Sets and Systems, 180, 146-163. https://doi.org/10.1016/j.fss.2010.11.009

21. Holcapek, M. (2011) A Smoothing Filter Based on Fuzzy Transform. Fuzzy Sets and Systems, 180, 69-97. https://doi.org/10.1016/j.fss.2011.05.028

22. Perfilieva, I., Novak, V. and Dvorak, A. (2008) Fuzzy Transforms in the Analysis of Data. International Journal of Approximate Reasoning, 40, 26-46. https://doi.org/10.1016/j.ijar.2007.06.003

23. Stěpnicka, M. and Polakovic, O. (2009) A Neural Network Approach to the Fuzzy Transform. Fuzzy Sets and Systems, 160, 1037-1047. https://doi.org/10.1016/j.fss.2008.11.029

24. Stěpnieka, M., Dvorák, A., Pavliska, V. and Vavrícková, L. (2011) A Linguistic Approach to Time Series Modeling with the Help of F-Transform. Fuzzy Sets and Systems, 180, 164-184. https://doi.org/10.1016/j.fss.2011.02.017

25. Troiano, L. and Ktriptani, P. (2011) Supporting Trading Strategies by Inverse Fuzzy Transform. Fuzzy Sets and Systems, 180, 121-145. https://doi.org/10.1016/j.fss.2011.05.004

26. Bede, B. and Rudas, I.J. (2011) Approximation Properties of Fuzzy Transforms. Fuzzy Sets and Systems, 180, 20-40. https://doi.org/10.1016/j.fss.2011.03.001

27. Perfilieva, I., Dankova, M. and Bede, B. (2011) Towards a Higher Degree F-Transform. Fuzzy Sets Systems, 180, 3-19. https://doi.org/10.1016/j.fss.2010.11.002

28. Di Martino, F., Loia, V. and Sessa, S. (2003) A Method for Coding/Decoding Images by Using Fuzzy Relation Equations. In: Bilgic, T., De Baets, B. and Kaynak, O., Eds., Fuzzy Sets and Systems—IFSA 2003, Lecture Notes in Artificial Intelligence, Vol. 2715, Springer, Berlin, 436-441. https://doi.org/10.1007/3-540-44967-1_52