Object Recognition Based On Sketch And Edge

Sketch is used to render the visual world since prehistoric times. Closely correlated with the increasing availability of digital touch-screen devices, re-search on human sketch object recognition attracts more attentions. Categoriz-ing free-hand human sketches has profound implications in applications such as sketch-based image retrieval (SBIR) and human computer interaction. Object detection is a fundamental problem in computer vison, and object proposals is regarded as a key preprocessing step for object detection nowadays due to it contributes significant speed-ups in object detection. In this thesis, we study the above critical problems including sketch generation for sketch-based image retrieval, sketch recogniton and object proposals.We propose a perceptual grouping framework that effectively solving the problem of sketch generation, which aims to produce human-drawing-like sketches from real photographs. In particular, we study how multiple Gestalt rules can be encapsulated into a unified perceptual grouping framework for sketch gener-ation. We further show that by solving the problem of Gestalt confliction, i.e. encoding the relative importance of each rule, more human-like sketches can be generated. For that, we release a manually labeled sketch dataset of 96 ob-ject categories and 7680 sketches. A novel evaluation framework is proposed to quantify human likeness of machine-generated sketches by examining how well they can be classified using models trained from human data. Afterwards, we demonstrate the superiority of our sketches under the practical application of sketch-based image retrieval.For the problem of sketch recognition, most prior works address this prob-lem following a standard supervised learning pipeline widely adopted for ob-ject recognition. One of the most fundamental problem is how to effectively describe a sketch image. We present a novel patch-based sparse representation (PSR) for describing sketch image and it is evaluated under a sketch recognition framework. It’s the first time that sparse coding is utilized for sketch recog-nition. Extensive experiments on a large scale human drawn sketch dataset demonstrate the effectiveness of the proposed descriptor.However, prior works on sketch recogniton often assume the availability of a large training set, rendering them sensitive towards abstraction and less scalable to new categories. To overcome this limitation, we propose a transfer learning framework which enables one-shot learning of sketch categories. The framework is based on a novel co-regularized sparse coding model which ex-ploits common/shareable parts among human sketches of seen categories and transfer them to unseen categories. Experimental results reveal that the pro-posed method offers a two-fold improvement over baselines.Finally, we improve the perceptual grouping framework that organizes im-age edges into meaningful structures and demonstrate its usefulness on object proposals. Our grouper formulates edge grouping as a graph partition problem, where a learning to rank method is developed to encode probabilities of candi-date edge pairs. Afterwards, an edge grouping based object proposal measure is introduced that yields proposals comparable to state-of-the-art alternatives.