A perceptually motivated online benchmark for image matting

Abstract

The availability of quantitative online benchmarks for low-level vision tasks such as stereo and optical flow has led to significant progress in the respective fields. This paper introduces such a benchmark for image matting. There are three key factors for a successful benchmarking system: (a) a challenging, high-quality ground truth test set; (b) an online evaluation repository that is dynamically updated with new results; (c) perceptually motivated error functions. Our new benchmark strives to meet all three criteria.We evaluated several matting methods with our benchmark and show that their performance varies depending on the error function. Also, our challenging test set reveals problems of existing algorithms, not reflected in previously reported results. We hope that our effort will lead to considerable progress in the field of image matting, and welcome the reader to visit our benchmark at www.alphamatting.com. 1826 978-1-4244-3991-1/09/$25.00 ©2009 IEEE

Improving Sampling Criterion for Alpha Matting

“…7 shows the comparison of the results between without post-processing and using guided filter as post-processing. We tested 27 benchmark images with the given trimaps provided by [13]. The average SAD error about four methods (i.e., Robust Matting [1], Shared Matting [4], Global Sampling Matting [3] and our method) is shown in Fig.…”

“…Besides, an effective cost function is applied to select samples in our work. We test the 27 images provided by [13] with the given trimaps and compare our work with some superior approaches. The results show that our method is competitive and effective to produce high-quality mattes.…”

“…6. (a) Images provided by[13]. (b) The estimated alpha mattes using our method with the given trimaps (without post-processing).…”

A comprehensive study on digital image matting

“…We test various combination state-of-the-art matting methods on the benchmark datasets [20]. There are 27 training images with groundtruth in total, then we compute the peak signal-to-noise ratio (PSNR) of each method for all these images.…”

“…3. Image matting results tested on benchmark data sets [20]. (a) input image, (b) closed-form matting [12], (c) KNN matting [14], (d) groundtruth alpha, (e) learning based matting [3] and (f) color clustering matting [13].…”