Very Low Cost Algorithms for Predicting the File Size of JPEG Images Subject to Changes of Quality Factor and Scaling
Data Compression Conference (2008)
Mar. 25, 2008 to Mar. 27, 2008
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/DCC.2008.85
This work presents two new algorithms to predict the file size of a JPEG image subject to transformations consisting of simultaneous changes in resolution (scaling) and in quality factor (QF). To be computationally efficient, the prediction is based solely on easily accessible image parameters such as the quality factor andthe original file size. A large image corpus (100 000 images), gathered by a crawler, is divided into a training set used to optimize the predictors and into a test set used to validate the predictors. For both algorithms the prediction error is shown to be of a few percents when the output parameters are close to those of the original image while remaining reasonably attractive elsewhere. Both algorithms are simple to implementand require very little processing for the prediction itself; making them good choices for implementation in transcoding servers.
Transcoding, image adaptation, resolutionTranscoding, image adaptation, resolution reduction, file size prediction, JPEG
S. Pigeon and S. Coulombe, "Very Low Cost Algorithms for Predicting the File Size of JPEG Images Subject to Changes of Quality Factor and Scaling," 2008 Data Compression Conference(DCC), Snowbird, UT, 2008, pp. 538.