公开(公告)号：US09547908B1

公开(公告)日：2017-01-17

申请号：US14868332

申请日：2015-09-28

Applicant: Google Inc.

Inventor： Eunyoung Kim ,  Ronald Frank Wotzlaw

IPC: G06K9/34 ,  G06T7/00 ,  G06T7/40 ,  G06T3/40 ,  G06T5/20 ,  G06K9/00

CPC classification number: G06K9/00228 ,  G06K9/00234 ,  G06K9/00362 ,  G06K9/469 ,  G06T5/20 ,  G06T7/11 ,  G06T7/162 ,  G06T7/194 ,  G06T2207/10024 ,  G06T2207/20028 ,  G06T2207/30196

Abstract: Implementations relate to feature mask determination for images. In some implementations, a computer-implemented method to determine a feature mask for an image includes estimating one or more prior regions in the image that define a feature in the image. The method determines superpixels based on multiple pixels of the image similar in color. The method constructs a graph, each node of the graph corresponding to a superpixel, and determines a superpixel score for each superpixel based on a number of pixels of the superpixel. The method determines one or more segmented regions in the image based on applying a graph cut technique to the graph based at least on the superpixel scores, and determines the feature mask based on the segmented regions. The feature mask indicates a degree to which pixels of the image depict the feature. The method modifies the image based on the feature mask.

Abstract translation: 实现涉及图像的特征掩模确定。 在一些实现中，用于确定图像的特征掩模的计算机实现的方法包括估计图像中定义图像中的特征的一个或多个先前区域。 该方法基于颜色相似的图像的多个像素来确定超像素。 该方法构成了与超像素相对应的图形的每个节点的图形，并且基于超像素的像素的数量确定每个超级像素的超像素分数。 该方法基于至少基于超像素得分将曲线切割技术应用于图形来确定图像中的一个或多个分割区域，并且基于分割区域确定特征掩模。 特征掩码表示图像的像素描绘特征的程度。 该方法基于特征掩码修改图像。