Parallelization of image similarity analysis
The algorithmical architecture and structure is presented for the parallelization of image similarity analysis, based on obtaining multiple digital signatures for each image, in which each "signature" is composed by the most representative coefficients of the wavelet transform of the corre...
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Articulo |
| Lenguaje: | Inglés |
| Publicado: |
2001
|
| Materias: | |
| Acceso en línea: | http://sedici.unlp.edu.ar/handle/10915/9421 http://journal.info.unlp.edu.ar/wp-content/uploads/p7.pdf |
| Aporte de: |
| Sumario: | The algorithmical architecture and structure is presented for the parallelization of image similarity analysis, based on obtaining multiple digital signatures for each image, in which each "signature" is composed by the most representative coefficients of the wavelet transform of the corresponding image area. In the present paper, image representation by wavelet transform coefficients is analyzed, as well as the convenience/necessity of using multiple coefficients for the study of similarity of images which may have transferred components, with change of sizes, color or texture. The complexity of the involved computation justifies parallelization, and the suggested solution constitutes a combination of a multiprocessors "pipelining", being each of them an homogeneous parallel architecture which obtains signature coefficients (wavelet). Partial reusability of computations for successive signatures makes these architectures pipelining compulsory. |
|---|