Showing papers by "Naoi Satoshi published in 2016"

Highlight removal for camera captured documents based on image stitching

Abstract: In camera captured document image analysis, specular highlights on glossy document surfaces often have negative effects on state-of-the-art Optical Character Recognition systems. Traditional highlight removal methods rely on either texture information from a single image or specific hardware setups to fuse complementary visual cues from multiple images. Motivated by the image stitching approaches to seamless panorama creation, we propose to remove specular highlights by capturing a pair of images from two viewing angles and merge their specular-free regions to produce an enhanced image eligible for character recognition. Compared to traditional solutions, our stitching based method can deal with document images with severe specular highlights, and eliminate the need for specific hardware devices or rigorous setup. Experimental results on various glossy document images demonstrate the effectiveness and efficiency of our method.

Training method and apparatus for neural network for image recognition

Abstract: A training method and a training apparatus for a neutral network for image recognition are provided. The method includes: representing a sample image as a point set in a high-dimensional space, a size of the high-dimensional space being a size of space domain of the sample image multiplied by a size of intensity domain of the sample image; generating a first random perturbation matrix having a same size as the high-dimensional space; smoothing the first random perturbation matrix; perturbing the point set in the high-dimensional space using the smoothed first random perturbation matrix to obtain a perturbed point set; and training the neutral network using the perturbed point set as a new sample. With the training method and the training apparatus for the neutral network for image recognition, classification performance of a conventional convolutional neural network is improved, thereby generating more training samples, reducing influence of overfitting, and enhancing generalization performance of the convolutional neural network.

Perspective correction using camera intrinsic parameters

Abstract: This paper presents a method for image perspective correction using camera intrinsic parameters. This method is based on two assumptions: a) the taken picture have a rectangle area, but didn't know the rectangle area's aspect ratio; b) the camera's intrinsic parameters should obtain by picture (Intrinsic parameters can easily obtain in iPhone or Android phone). The rectangle's perspective distortion is constrained by camera's intrinsic parameters. So this paper's method find this constraint to get the distorted rectangle's aspect ratio. Then using this ratio can get the homography matrix. This method also can be used to detect card or document rectangle area under complex background. We took some pictures by iPhones and android phones, the results shows that our method can correct document pictures' perspective distortion.