ABSTRACT
Biometric fingerprint images require substantial storage, transmission and computation costs, thus their compression is advantageous to reduce these requirements. This research work presents a novel approach to biometric fingerprint image compression by the innovative application of non-uniform quantization scheme in combination with level-dependent threshold strategy applied to wavelet transformation as opposed to the widely used uniform quantization scheme. Comparative analysis of Coiflet wavelets implemented with level dependent thresholds and Daubechies wavelets were conducted on the basis of percentage retained energy, RE (%). The RE (%) values for Coiflet wavelet ranged from 99.32% to 99.69% as opposed to the values for Daubechies wavelet which ranged from 98.45% to 99.15%. These results revealed that the Coiflet wavelet bases performed better than the Daubechies wavelet. Hence, the choice of Coiflet wavelet for image transformation in the proposed compression algorithm was justified. The performance analysis of uniform and nonuniform scalar quantization schemes for biometric fingerprint image compression was conducted. The non-uniform quantization method based on Lloyd-Max approach performed better than the uniform quantization method used in the existing fingerprint compression standards. The Signal-to-Quantization Noise Ratio (SQNR) values for non-uniform quantization increased from 19.2977 dB for 3 bit per pixel (bpp) to 44.6083 dB for 7 bpp whereas for the same range (3 bpp to 7 bpp) for uniform quantization, SQNR values increased from 17.0903 dB to 40.1349 dB. Therefore, non-uniform quantization based on Lloyd-Max approach was employed for this compression algorithm. The implementation of the proposed biometric fingerprint image compression algorithm involved three stages, namely: the transformation of biometric fingerprint image; non-uniform quantization of transformed image and the entropy coding which is the final stage. In order to determine the overall performance of the algorithm, Peak Signal-to-Noise Ratio (PSNR) and Compression Ratio (CR) were used as performance metrics. PSNR was used as a measure of the resultant image quality after compression and the Compression Ratio was used as a measure of the degree of compression achievable. A trade-off was made between the achievable compression ratio and the realizable image quality which is a function of the achievable PSNR in the overall compression process. The overall performance of the proposed compression algorithm achieved an improvement in terms of compression ratio of 20:1 over the existing compression standard for biometric applications which have a compression ratio limit of 15:1. The improvement was largely due to the novel approach employed in this research work as stated above.