Abstract: Image fusion is the process of merging two different types of images for the same region to get one image. There are different techniques for image fusion like Intensity Hue Saturation (IHS), Principal Component Analysis (PCA), High Pass Filter (HPF). The raw images and the fused image usually require a large storage to save them, so it will be difficult to save them on the Internet. Here is the aim of image compression is to reduce the size of the image to be able to save it on the Internet, so This Research is talking about the effect of lossy compression on High Pass Filter image fusion.
Keywords: lossy compression, MrSid, image fusion, HPF, RMSE
1. Introduction:
From remote sensing, we can get two types of images. One of the them is called the panchromatic image (pan) with high spatial resolution and low spectral resolution, and the other one is called the multispectral image (mul) with high spectral resolution and low spatial resolution. The images that were used in this research are for Cape Town, South Africa. The pixel of the pan image is 0.5m*0.5m and the pixel of the mul image is 2.0m*2.0m. The image fusion is the process to merge these two images into one image with high spatial resolution and high spectral resolution. The method of image fusion used in this research is high pass filter (HPF). After that we compressed the two raw images with different compression ratio (1:10, 1:20, 1:30, ………. 1:100) by using the Discrete Wavelet Transformation (MrSid format), and then we fused the compressed raw images to get 10 fused images. Finally, we compared between the fused image that came from the two raw images and the 10 fused images those came from the compressed raw images with different compression ratios.
2. Related work:
2.1. Image fusion:
2.1.1. High Pass Filter (HPF):
A high-pass filter (HPF) passes high-frequency signals, as it is an electronic filter, but it lows the amplitude of signals with frequencies lower than the cutoff frequency. The actual quantity of attenuation for each frequency has variation from filter to filter. A high-pass filter is usually represented as a linear time-invariant system [1].
At first, it is computed the ratio between the panchromatic high spatial resolution image and the multispectral image [2]. A high pass convolution filter kernel is formed and it is used based on the ratio to filter the panchromatic high spatial resolution input data with the size of the kernel. The HPF image is supplemented to each band of the multispectral image. After the HPF image is weighted relative to the global standard deviation of the multispectral bands with the weight issues again computed from the ratio, the summation is done [3]. In the final step, a linear stretch is done to the new multispectral image to identify the standard deviation and the mean values of the original input multispectral image. It shows the good and acceptable results also for multi temporal and multi sensorial data. The edges are sometimes stressed too much.
2.2. Image compression:
2.2.1 Image compression based on Discrete Wavelet Transformation (MrSid format):
In the remote sensing field, because of the availability of high-resolution and hyper-spectral satellite imagery, and the panchromatic image and the multispectral image need large space to save the images on the internet, image compression has become more significant. [4]. The Spatial Data Infrastructures (SDI) paradigm improved over the years, develops the establishment of web data services, usually in terms of Open Geospatial Consortium (OGC) standards like the Web Map Service (WMS) and the Web Coverage Service (WCS). Nevertheless, it is very important to do interactive transmission strategies and compression for the images in these web services, so it could be easy to save and transfer very large images to environments with restricted bandwidth [5].
There are two broad types of image compression methods (lossy and lossless compression). Lossy and lossless compression methods are terms that illustrate whether or not if the original image can lose any data when the compressed one is uncompressed [6].
Lossless compression consists of methods, which completely keeps the original data without losing data. When the compressed image is decompressed back to the original image, it is numerically match with the original image, and is therefore favored in life critical situation, such as the archival of business documents or medical image, etc. where losing any date in the image and quality could cause incorrect analysis and that has a sever adverse influence [7]. There must be some redundancy in the original data for this type of compression to be efficient.
The other type of image compression is lossy compression. It works on minimizing an image size by permanently eliminating certain information, especially redundant information. After uncompressing the image, only a part of the original information is still there. This type of compression is chosen in different applications because of giving high compression ratio that results in smaller image sizes needing less space however the spatial and spectral features of the image are lost. Even though in some cases the visual impact of a lossy technique may be imperceptible [8][9].
Multiresolution Seamless Image Database (MrSID) is a compression algorithm based an wavelet transform. It was designed by LizardTech. The story of improvements in MrSID includes a memory effective implementation and automatic inclusion of pyramid layers in every data set, both of which make MrSID is very suitable to give effective retrieval of very large digital images and effective storage. The methodology of the underlying wavelet-based compression that was used in MrSID yields high compression ratios while satisfying stringent image quality requirements. The compression method that is used in MrSID is the type of lossy compression ( the process of compression and decompression losses some of the source data pixel-for-pixel). The compression methods that are used by MrSID technology give both high performance and high quality imagery while still meeting our industry’s challenging workflows. [10][11]
2.3. Evaluating fusion methods:
Evaluating fusion methods is depend on the verification of the preservation of spectral characteristics and the development of the spatial resolution. The fused images are compared visually. The visual appearance depends on the human’s eye to interpret the fused image and it may be subjective, but a number of statistical evaluation techniques are used to compare between the different fusion methods and to measure the color because the power of the visual cognition as a final backdrop cannot be underestimated. These methods should be objective, reproducible, and of quantitative nature. The RMSE is calculated as the difference between the mean and the standard deviation of the fused image and the original image. The best value is to be equal zero. The following equation shows how to calculate the RMSE:
RMSE=√((〖std〗_i-〖std〗_f )^2+(〖mean〗_i-〖mean〗_f )^2 ) …………………….. (1)
3. Results and analysis:
we compared the fused image came from the decompressed raw pan and mul images with the 10 fused images came from the compressed pan and mul images with different compression ratios to get the RMSE to study the effect of compression on image fusion, as shown in table1.
Compression ratio “1:10” “1:20” “1:30” “1:40” “1:50” “1:60” “1:70” “1:80” “1:90” “1:100”
RMSE 0.339054 1.220939 2.517272 3.247004 4.847105 5.286009 6.847213 7.26097 7.954781 8.019851
Table 1 illustrates the RMSE for the fused image came from the raw images and the 10 fused images came from the compressed images
From chart 1, it is obvious that compression ratio 1:10 gives the least RMSE as the RMSE equals 0.339 and the RMSE increases as the compression ratio increases from 1:20 to 1:100. The compression ratio 1:100 gives the highest RMSE.
Chart 1 illustrates the RMSE for the fused image came from the raw images and the 10 fused images came from the compressed images
4. Conclusion:
From the results it was found that the compression ratio 1:10 is the best and the perfect ratio that has a little effect on image fusion as it gives the lowest RMSE and the RMSE equals 0.339, while the other compression ratios (1:20, 1:30, ….. 1:100) give higher RMSE as it increases by increasing the compression ratios from 1:20 to 1:100 that gives the highest RMSE and the compression ratio 1:100 has the worst effect on the image fusion. We can make use of that in many applications of remote sensing. For example, we can make use of that in classification of palins, hills, and mountains as the best results for kappa coefficient when the compression ratio is less than 1:10 [12]. For the classification of crops and forests, the best compression ratio is from 1:3.33 to 1:5.
5. References:
- Watkinson, John (1998). The Art of Sound Reproduction. Focal Press. pp. 268, 479. ISBN 0-240-51512-9. Retrieved March 9, 2010.
- Main, Bruce (February 16, 2010). “Cut ‘Em Off At The Pass: Effective Uses Of High-Pass Filtering”. Live Sound International (Framingham, Massachusetts: ProSoundWeb, EH Publishing). Figure 1 illustrates the original images (a) multispectral image, and (b) panchromatic image / Figure 2 illustrates the HPF fused image form the raw pan and mul images
- Paul M. Mather (2004). Computer processing of remotely sensed images: an introduction (3rd ed.). John Wiley and Sons. p. 181. ISBN 978-0-470-84919-4.
- Lau, W.L., Li, Z.L. and Lam, K. W., 2003, Effects of JPEG compression on image classification, International Journal of Remote Sensing, Vol. 24, No.7, pp. 1535-1544.
- Zabala, A. and Pons, X., 2011, Effects of lossy compression on remote sensing image classification of forest areas, International Journal of Applied Earth Observation and Geoinformation, Vol. 13, pp. 43–51
- Zhai, L., Tang, X. M., Zhang, G., and Wu, X., 2008, Effects of JPEG2000 and SPIHT compression on image classification, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Beijing, China, Vol. XXXVII, Part B7, pp. 541–544.
- M. Mozammel Hoque Chowdhury and Amina Khatun, 2012, Image Compression Using Discrete Wavelet Transform, (IJCSI) International Journal of Computer Science Issues, Vol. 9, Issue 4, No. 1, pp. 327 – 330
- Shrestha, B., O’Hara, G. C., and Younan, H. N., 2005, JPEG2000: Image quality metrics, ASPRS 2005 Annual Conference, Geospatial Goes Global: From Your Neighborhood to the Whole Planet, Baltimore, Maryland.
- Campbell, J.B., 1996, Introduction to Remote Sensing, 2nd ed., Guilford, New York.
- ERDAS, 2010, ERDAS Field Guide, ERDAS, Inc., Norcross, GA.
- LizardTech, 2010, LizardTech’s MrSID Technology, Celartem Inc. dba LizardTech, Seattle, Washington, USA.
- Zhai Liang and Tang Xinminga , a Key Laboratory of Geo-Informatics of State Bureau of Surveying and Mapping, Chinese Academy of Surveying and Mapping, China
Update 2023
Since the essay was written in 2016, there have been several developments in the field of lossy image compression that are worth considering:
Deep Learning-based compression: One of the major developments in lossy image compression has been the use of deep learning-based algorithms for compression. These algorithms use neural networks to learn the most efficient way to represent an image using fewer bits. These methods have shown to produce high-quality compressed images while achieving higher compression rates compared to traditional compression algorithms.
Advances in hardware: Advances in hardware, such as faster processors and more efficient memory, have made it possible to use more complex algorithms for compression without sacrificing speed. This has enabled the development of new compression methods that can produce higher quality compressed images.
New image formats: New image formats such as AVIF (AV1 Image File Format) have been developed that offer better compression than existing formats like JPEG while maintaining similar image quality. AVIF is a new open-source image format that uses the AV1 video codec to compress images. This format has shown to provide better compression than other formats such as JPEG, PNG, and WebP, while maintaining similar or better image quality.
Perceptual-based compression: New compression methods based on perceptual quality metrics have been developed that aim to preserve the perceptual quality of the image. These methods take into account the human visual system’s response to image distortions and prioritize the preservation of the most important visual information.
In summary, recent advancements in lossy image compression have led to the development of new and improved compression methods that can achieve higher compression rates while maintaining or improving image quality. Deep learning-based compression algorithms, advances in hardware, new image formats, and perceptual-based compression methods are some of the key developments in this area.