Concept-based image search is an emerging search paradigm that utilizes a set of concepts as intermediate semantic descriptors of images to bridge the semantic gap.
Concept-based image search:基于概念的图像检索 。lso variably named as "description-based" or "text-based" image indexing/retrieval, refers to retrieval from text-based indexing of images that may employ keywords, subject headings, captions, or natural language text。还有一种是content based image retrival(CBIR),基于内容的图像检索,指利用图像查询图像,基于内容的图像检索指的是查询条件本身就是一个图像,或者是对于图像内容的描述,它建立索引的方式是通过提取底层特征,然后通过计算比较这些特征和查询条件之间的距离,来决定两个图片的相似程度。
bridge semantic gap:弥合语义鸿沟。在传统的基于文字的查询技术中,不存在这个问题,因为查询关键字基本能够反映查询意图。但是在基于内容的图像查询中,就存在一个底层特征和上层理解之间的差异(这也就是著名的semantic gap)。主要原因是底层特征不能完全反映或者匹配查询意图,如一幅关于节日的图像所表达出的欢乐和喜庆的感觉等,更需要根据人的知识来判断。
In concept-based image search, a set of concept detectors are prebuilt to predict the presence of specific concepts, which provide direct access to the semantic content of images. Given a textual query, it is mapped to a group of primitive concepts, and the search results are made up of the images in which these concepts are likely to appear。
现在visual concept detectioncurrent concept-based search techniques can effectively deal with queries involving only one concept.
不能简单通过联合几个concept来得到查询结果的理由有:
1 不同concept的权重可能不同,需要combination weigthts
2 各concept是相互作用的,有内在的语义
3 the information cues conveyed by the non-query concepts have not been fully exploited in prior concept-based image search methods.
例入concept detector得到了sofa 我们的查询包括了street ,那这个image应当是irrelevant。
本文的框架:
建立concept detector → 建立concept直接的相关性,即takes into account concept weights and concept correlations
→ ranking-oriented learning is ultimately developed to determine the model parameters through optimizing the image ranking performance for complex queries.
pairwise learning
ranking svm
learning to rank
factorization machine
model paradigm
explicitly model :显示建模
heuristically:启发式
hinge loss function :合页损失函数
data-driven approach:只要有足够代表性的样本(数据),就可以运用数学找到一个或者一组模型的组合使得它和真实的情况非常接近。注意这个方法的前提是具有「足够代表性的数据」。这种方法称为「数据驱动方法」