Research Document Imaging Image Processing Feature-based Filtering

Search

Feature-based Filtering for line-drawings
Dr. Eugene Ageenko

Intro

The feature-based filtering technique is based on the semantic image modeling that utilizes the global spatial dependencies in the image. Line-drawings consist mainly of straight-line elements, and global information can be gathered by extracting line features. The filtering is applied as a part of an image compression system. The feature extraction and the filtering are considered as preprocessing steps before the compression. The noise removal improves the image quality and alleviates the loss in the compression ratio caused by noise.

Filter in work

The feature-based filtering technique is based on the idea of flipping of isolated pixel groups found in the difference (mismatch) between the original image and one that is reconstructed from extracted vector features.

Input image	Output image	Filtered pixels

Evaluation and comparison

Figure. Comparison of the filtering methods used together with two compression standards. The figure shows the numbers as the relative reduction in file size when compressing the filtered images from the entire test set.

Conclusion

The proposed feature-based filtering technique removes additive and quantization noise from the original image, restores image quality, and in this way produces a better compression performance. For a set of test images, the method improves the compression ratio by about 17 % in comparison to JBIG1. The method can be considered as a preprocessing step to existing compression techniques, and standard decompression routines can be applied