| 作成者 |
|
|
|
|
|
|
|
| 本文言語 |
|
| 発行日 |
|
| 収録物名 |
|
| 巻 |
|
| 開始ページ |
|
| 終了ページ |
|
| 会議情報 |
|
| 出版タイプ |
|
| アクセス権 |
|
| 権利関係 |
|
|
|
| 関連DOI |
|
| 関連ISBN |
|
| 関連HDL |
|
| 概要 |
We propose a method for anonymizing a given image by removing personally identifiable information with a superpixel segmentation technique. Especially, we would like to focus on human faces in an imag...e, because face images bear important personal information. First, human faces in an image are detected by a Haar feature-based cascade classifier. Next, the detected regions are segmented by an iterative raster scan algorithm for superpixel segmentation. This algorithm produces compact and adherent superpixels, and has a smaller number of parameters than the simple linear iterative clustering (SLIC) which is a well-known superpixel segmentation algorithm. Since the obtained superpixels adhere well to image boundaries, the outlines of detected faces are preserved while the details are removed by averaging the colors in each superpixel for privacy protection. We experimentally demonstrate the effectiveness of the proposed method with example images which contain human faces, where the proposed method is compared with conventional methods for image anonymization based on blurring, masking and mosaicing effects.続きを見る
|
Mendeley出力
