Digital Image Processing

S. Batmavady ¹, M. Tamilarasi ², K. Manivannan ³, D. N. V. Sukesh ³

Affiliations
1 Department of ECE, Pondicherry Engineering College, Pillaichavady, Puducherry, IN
2 Dept. of ECE, Pondicherry Engineering College, Pillaichavady, Puducherry, IN
3 Pondicherry Engineering College, IN

Abstract

Image fusion techniques are applied in various fields such as remote sensing, medical imaging, concealed weapon detection, etc. Combining two or more images of the same scene usually produces an output image which provides increased interpretation capabilities and reliable results. In image fusion, data with different specifications such as resolution, spectral and spatial coordinates are combined. Image fusion algorithm can be categorized into pixel and feature levels. Region based method is one way of achieving the feature- level fusion. Segmentation plays a vital role in this fusion process where the features of the source images are extracted first using Edge based segmentation Consequently, the Contourlet transform is applied on the different regions and the coefficients from different regions are merged separately. Finally, the fused image is obtained by performing inverse Contourlet transform. The Laplacian pyramid employed in Contourlet transform is not the perfect transform from the point of view of image fusion, since it involves down-sampling procedure which makes it shift variant. Therefore, in order to yield better performance metric in the proposed work, the Contourlet transform is modified by replacing the Laplacian pyramid by Contrast pyramid. Region based image fusion using modified Contourlet transform and the Contourlet transform are applied on various images to compare their performances. Simulation results indicate that Region Based Image Fusion using Modified Contourlet transform produces better results than Contourlet transform in terms of entropy, correlation coefficient, PSNR and average gradient.

Keywords

Contrast Pyramid, Directional Filter Bank, Image Fusion, Modified Contourlet Transform, Segmentation.

Full Text

     

S. Batmavady ¹, H. Rekha ², K. Manivannan ³

Affiliations
1 ECE Dept., Pondicherry Engg. College, Pillaichavady, Puducherry – 605014, IN
2 ECE Dept, Pondicherry Engg. College, Puducherry–605014, IN
3 EEE Dept. of Pondicherry Engg. College, Puducherry – 605 014, IN

Abstract

In this paper, a novel region–based image fusion technique using different wavelet transforms, that integrates multiscale image segmentation and a statistical fusion approach is considered for fusing the multisensor images. Compared to pixel-level image fusion schemes, region-based fusion schemes are less sensitive to noise. But, the region based wavelet transform technique is also vulnerable to noise, as it classifies all noise to new regions with different frequency bands. The effect of noise in the image can however be suppressed using advanced wavelet transform like dual tree complex wavelet transform and dual tree complex wavelet packet transform, which possess properties like shift invariance and directionality. In some cases, the frequency decomposition provided for the signals by the DTCWT might also be not optimal. This drawback is overcome in this paper by using the dual tree complex wavelet packet transform (DTCWPT) which provides good directionality, shift invariance and better image denoising. Performance comparison using quantitative measures like PSNR, entropy and RMSE indicates the effectiveness of the proposed method over other techniques.

Keywords

DTCWT, DTCWPT, Image Fusion, Multisensor Images, Segmentation.

Full Text

     

S. Batmavady ¹, R. Senthil ¹, K. Manivannan ²

Affiliations
1 Dept. of ECE, Pondicherry Engg. College, Pillaichavady, Puducherry-605014, IN
2 EEE Dept., Pondicherry Engg. College, Pillaichavady, Puducherry-605014, IN

Abstract

The proposed paper presents a new approach to color image segmentation by fusing Fuzzy C-Means (FCM) and Atanassov's Intuitionistic Fuzzy Set (A-IFS) Histon. The method proposed here does not require prior knowledge about the no. of clusters. The proposed method also calculates the threshold value automatically for segmentation. This method calculates membership values using conventional FCM and uses A-IFS Histon for calculating the no. of clusters and employs roughness index for segmentation. In the homogeneous region, the roughness index is more and is close to unity value, whereas, in the boundary region, it is small. The qualitative and quantitative comparison of the proposed method yields better results against the recently developed A-IFS Histon technique.

Keywords

Atanassov's Intuitionistic Fuzzy Set, Clustering, Color Image Segmentation, FCM, Membership Values.

Full Text