No abstract

A new license plate detection method for challenging environments is proposed. Background clutters are common in road scene images and the detection of license plates (occupying only a small part of an image) is considered as a difficult problem. In order to address this problem, a two‐step approach is developed: first vehicle regions are detected and the license plate in each vehicle region is localised. This vehicle region detection based approach provides scale information and limits search ranges in license plate detection, so that one can reliably detect license plate regions. To be precise, the faster region‐based convolutional neural network algorithm for the vehicle region detection is adopted and candidates for license plates in each detected region with the hierarchical sampling method are generated. Finally, non‐plate candidates are filtered out by training a deep convolutional neural network. The proposed method is evaluated on the Caltech dataset and the method showed a precision of 98.39% and a recall of 96.83%, which outperforms conventional methods.

No abstract

No abstract

No abstract

This paper proposes a new approach to the estimation of document states such as interline spacing and text line orientation, which facilitates a number of tasks in document image processing. The proposed method can be applied to spatially varying states as well as invariant ones, so that general cases including images of complex layout, cameracaptured images, and handwritten ones can also be handled. Specifically, we find CCs (Connected Components) in a document image and assign a state to each of them. Then the states of CCs are estimated using an energy minimization framework, where the cost function is designed based on frequency domain analysis and minimized via graph-cuts. Using the estimated states, we also develop a new algorithm that performs text block identification and text line extraction. Roughly speaking, we can segment an image into text blocks by cutting the distant connections among the CCs (compared to the estimated interline spacing), and we can group the CCs into text lines using a bottom-up grouping along the estimated text line orientation. Experimental results on a variety of document images show that our method is efficient and provides promising results in several document image processing tasks.

No abstract

No abstract