Proposal for multispectral image compression methods

Abstract

ABSTRACTThis paper has proposed two image compression and decompression schemes for multispectralimages. Two issues were considered in the proposed methods. The first issue is the possibility ofapplying the compression process directly to a set of multispectral images, where the standardJPEG should be applied to each individual image. Considering this issue, a compression anddecompression method is proposed based on a hybrid of lower bit suppresion and Karhunen-Loeve transform and named as KLT Hybrid. The second issue is the possibility of obtaining ageneral codebook for a bulky of typical data such as a set of hyperspectral images. Consideringthis issue, another compression and decompression method is proposed based on vectorquantization (VQ) where the general codebook is obtained by a proposed fair-share amountmethod. Four performance indicators were used to evaluate the results. The indicators includecompression ratio, root mean square error, maximum absolute error, and signal to noise ratio. Theexperimental results have shown good performance indication of both methods.Keywords: multispectral, hyperspectral, spectrometer, image compression and decompression,JPEG, transform coding, vector quantization1. INTRODUCTIONMultispectral images are large in size and consist of redundancies. To be able to send the data viacommunication channels, and to enhance the data processing time, and to store the dataefficiently, the data should be compressed. The trend of multispectral system that increases thenumber of its channels has made the multispectral image compression a recent topic issue [1-3].JPEG (Joint Photographic Expert Group) has been accepted as one of the standard imagecompression method. JPEG compression method is based on Direct Cosine Transformation(DCT) and quantization process [4]. This approach is useful especially in the application ofteleconferencing or document transmission where people still can recognize the objects containedin the image even when loosing the details. The approach has taken advantage of the weaknessesof the human eyes in capturing the detail of gray level intensities. Besides that, the use ofstandard JPEG method for multispectral image compression should be done to each individualimage and it will be cumbersome if the multispectral system has a high number of channels, suchas in hyperspectral system or spectrometer system.Recently, researchers use either the approach of transform coding [1,2] or vector quantization(VQ) [3]. The important issues in the two approach is how to apply the compression processdirectly to the set of images instead of to each individual image, and how to generate a generalcode book for all images in the set. Hybrid approach is commonly used in transform-codingmethod, such as hybrid of Difference Pulse Code Modulation and Direct Cosine Transform in [5]and hybrid of Direct Cosine Transform and quantization in [4]. The data reduction technique thatis usually applied for multispectral images is the Karhunen-Loeve Transform (KLT).