Use of Affine Invariants in Locally Likely Arrangement Hashing for Camera-Based Document Image Retrieval

Abstract

Abstract. Camera-based document image retrieval is a task of searching document images from the database based on query images captured using digital cameras. For this task, it is required to solve the problem of "perspective distortion" of images, as well as to establish a way of matching document images efficiently. To solve these problems we have proposed a method called Locally Likely Arrangement Hashing (LLAH) which is characterized by both the use of a perspective invariant to cope with the distortion and the efficiency: LLAH only requires Ç´AEµ time where AE is the number of feature points that describe the query image. In this paper, we introduce into LLAH an affine invariant instead of the perspective invariant so as to improve its adjustability. Experimental results show that the use of the affine invariant enables us to improve either the accuracy from 96.2% to 97.8%, or the retrieval time from 112 msec./query to 75 msec./query by selecting parameters of processing.

Toward augmenting everything: Detecting and tracking geometrical features on planar objects

“…Because the result was influenced by both the minimum distance between keypoints and the maximum number of extracted keypoints, we need to set up two thresholds considering an image size and content. To be invariant to viewpoint changes, one possible approach is to prepare synthetic images of various different viewpoints such as [22] and compute descriptors on them because the number of registered images did not much influence the retrieval costs due to a hashing scheme as reported in [21].…”

Geometrically-Correct Projection-Based Texture Mapping onto a Deformable Object

“…A major feature is the marker does not require a square black frame. Using Local Likely Arrangement Hashing (LLAH) [16], they utilize the local patterns of the dots for descriptors in keypoint matching and tracking. Leveraging the fact LLAH works well on the unique local patterns of keypoints, they make the assumption the randomness of the distribution will lead to the unique patterns in the dots.…”

“…We search for the reference pattern to find the most similar region. A location on the reference pattern can be uniquely recognized among 2 16 /4(U p, down, le f t, right) = 16384 (types of patterns) as long as any 4 × 4 rectangle pattern is observed by a camera. The matching with the reference pattern is conducted for every 4 × 4 units among the created grids.…”

Shared Annotations: The Social Side of Exam Preparation

“…The server receives and processes them, using a computer vision procedure, in order to separate the annotations from the original document. More technically each shared page/slide is first recognized using locally likely arrangement hashing [23]. Second, it is rectified based on the estimated planar pose [9].…”