Pre-processing is an important step in the field of image processing. The representation and quality of image is the first and foremost before running the analysis. Raw data is highly susceptible to noise and inconsistency. In order to help improve the quality of the data and, consequently, of the results raw data is pre-processed so as to improve the efficiency and ease of processing. In the context of image processing, it improves the ease of encoding at the source.If there is much irrelevant and redundant information present or noisy and unreliable data, then knowledge discovery during the training phase is more difficult. The preparation and filtering steps can take considerable amount of processing time. Image pre-processing involves cleaning, normalization, transformation, feature extraction and selection, etc. The product of image pre-processing is the final training set.
The following schemes are implemented and compared:
- Singular Value Decomposition
- Compressed Sensing
- Integer Wavelet Transform
The figure below shows the influence of principal components in PSNR for SVD implementation.
The figure below shows the relation between PSNR, compression ratio and sparsity of image for Compressed Sensing implementation.
The figure below shows IWT results.
Among the pre-processing techniques considered, IWT proves to be the best, followed by SVD (in terms of visual quality). Compressed sensing is effective for sparse images. As sparsity decreases, efficiency of Compressed sensing decreases.Assuming ideal channel conditions (channel noise = 0), IWT gives perfect reconstruction.