儿童照合成研究与非接触式掌纹定位方法
文献类型：学位论文
| 作者 | 林立 |
| 学位类别 | 工学硕士 |
| 答辩日期 | 2013-06-02 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 刘迎建 ; 黄磊 |
| 关键词 | 儿童照合成 人脸变形 掌纹定位 轮廓提取 Child picture synthesis Face morphing Palmprint localization Contour extraction |
| 其他题名 | Research of Child Picture Synthesis and Contactless Palmprint Localization Method |
| 学位专业 | 计算机应用技术 |
| 中文摘要 | 随着人们生活水平的不断提高和信息领域的迅速发展，生物特征领域得到人们越来越多的关注和研究。一方面，生物特征比如人脸由于其普遍性，在娱乐方面的应用得到了广泛关注。儿童照片合成 是人脸娱乐的一种，它是在给定一幅男性图片、一张女性图片的基础上，利用内置的儿童模板库，进行的图像融合技术。另外一方面，生物特征识别作为一种较为新颖的身份鉴别方式，得到了极为广泛的研究和应用。非接触式掌纹识别是生物特征识别领域一位新兴的成员，它通过分析掌纹内部的纹线结构来进行身份的认证和鉴别，在本文中，我们聚焦于非接触式掌纹识别的关键步骤-掌纹定位。 本文围绕儿童照片的合成和非接触式掌纹定位，开展了下述研究工作: (1) 本文提出一种儿童照合成的方法。分别给定男性、女性和原始儿童的图片，该方法从人脸形状和肤色两个层面上融合生成目标儿童图片。 对于人脸形状，直接采用线性组合的方法完成人脸形状的融合。对于人脸肤色，先采用高斯径向基变换进行人脸的变形，再利用交叉融合的方法完成人脸肤色的融合。实验证明，利用本文方法融合得到的儿童不仅具有父母亲的特征，而且具有比较高的视觉美感。 (2) 在掌纹定位中，针对复杂场景下掌纹关键点难以定位的问题，提出了一种由粗到细的掌纹关键点定位算法。先利用统计学习方法进行关键点的粗定位，然后利用一条最大能量路径限制未知关键点的位置，最后在路径周围重新定位出所有的关键点。 该算法不仅对复杂环境有着较好的适应性，对手掌的张开闭合也有着良好的鲁棒性。 (3) 针对掌纹ROI定位不稳定的问题，本文提出一种新的掌纹ROI获取方法。首先，在掌纹关键点的基础上，提出一种掌纹边缘轮廓提取算法。然后，结合掌纹边缘轮廓和掌纹关键点提取掌纹ROI。实验证明，该方法能够准确地提取掌纹ROI，并且具有较高稳定性。 |
| 英文摘要 | With the improvement of people's living standard and rapid development of information industry, biometrics field attracts more and more attention and study. On the one hand, biometrics such as face are widely applied in entertainment because of their universalities. Child picture synthesis is one popular application of face entertainment.It generates a new child picture by providing a man's picture、a woman's picture and the inside database of children pictures.On the other hand, biometrics as the novel authentication way have been widely studied and applied. Contactless palmprint recognition is a novel member of biometrics. It finishes authentication by analyzing line-like textures inside central regions of palmprint. In this thesis, we focus on the palmprint localization the key stage of contactless palmprint recognition. In this dissertation, we focus on child picture synthesis and contactless palmprint localization. The main contributions of this thesis include: (1) In this thesis, we propose one child picture synthesis method. Given a man's picture、a woman's picture、and a child's picture, the method generates the target child picture by blending these facial shapes and skin colors. We first get the target child's shape by direct linear combination. To get the target child's skin color, we finish the face warping via the Gaussian radical base function, then adopt the cross-dissolves method for the child picture synthesis. Experimental results indicate that the child face generated with our method not only owns the parents' features, but the picture also shows visual aesthetic feeling。 (2) In palmprint localization, key points are difficult to detect in the complex scenes. We propose a coarse-to-fine algorithm for key points detection.At first, the key points are coarsely detected by statistical learning based method, then, a max energy path is obtained to constrain the position of key points, at last, the key points can be accurately located around the path. Experimental results show that the proposed method has a good tolerance to complex scenes and different hand's open-closed degrees. (3) Considering the instability of traditional ROI(region of interest) acquisition methods, the thesis proposes a novel ROI acquisition method. At first, based on the key points localization,we propose a palmprint contour extraction method to obtain two contours of palm, then acquire the ROI by combining the information of palmprint contours and key points.... |
| 语种 | 中文 |
| 其他标识符 | 201028014629074 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/7694]  |
| 专题 | 毕业生_硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 林立. 儿童照合成研究与非接触式掌纹定位方法[D]. 中国科学院自动化研究所. 中国科学院大学. 2013. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。