Off-line Segmentation Research Articles

Offline segmentation of spatio-temporal order trajectories by mixed-integer linear programming for determining process times in production systems

Real-time indoor positioning systems in manufacturing systems are used to track production orders. This generates spatio-temporal trajectories which can be segmented to determine process times. We present formulations of the offline segmentation problem as mixed-integer linear programs (MILPs) that utilize the sequence of processing steps from ERP systems. The MILP formulations are compared with online heuristics in terms of their accuracy and computational effort on data generated with features from a real job shop. We show that in terms of accuracy our offline segmentation formulations outperform the online heuristics with increasing measurement errors, justifying their higher computational effort.

Contrasting off-line segmentation decisions with on-line word segmentation during reading.

In two experiments, we investigated the correspondences between off‐line word segmentation and on‐line segmentation processing during Chinese reading. In Experiment 1, participants were asked to read sentences which contained critical four‐character strings, and then, they were required to segment the same sentences into words in a later off‐line word segmentation task. For each item, participants were split into 1‐word segmenters (who segmented four‐character strings as a single word) and 2‐word segmenters (who segmented four‐character strings as 2 two‐character words). Thus, we split participants into two groups (1‐word segmenters and 2‐word segmenters) according to their off‐line segmentation bias. The data analysis showed no reliable group effect on all the measures. In order to avoid the heterogeneity of participants and stimuli in Experiment 1, two groups of participants (1‐word segmenters and 2‐word segmenters) and three types of critical four‐character string (1‐word strings, ambiguous strings, and 2‐word strings) were identified in a norming study in Experiment 2. Participants were required to read sentences containing these critical strings. There was no reliable group effect in Experiment 2, as was the case in Experiment 1. However, in Experiment 2, participants spent less time and made fewer fixations on 1‐word strings compared to ambiguous and 2‐word strings. These results indicate that the off‐line word segmentation preferences do not necessarily reflect on‐line word segmentation processing during Chinese reading and that Chinese readers exhibit flexibility such that word, or multiple constituent, segmentation commitments are made on‐line.

Offline segmentation and script recognition of Hindi using knowledge based approach and multi layered perceptron neural network

The following paper focusses on a segmentation technique for most widely used language in India i.e. Hindi. Because of variations in writing styles and intricate structures, this field is drawing a lot of attention from researchers. This paper proposes a system which mines and identifies offline Hindi words using MLP Neural Network. Here we are using three layer architecture of neural networks that consists of a layer for Input, Hidden layer and a Resultant layer. Also segmentation is done using Knowledge Based Approach. The success rate after segmentation is 94.87%. Also the overall recognition accuracy is 85.78%.

Partitioning continuous segmented signals

An off-line segmentation of a continuous-time signal is proposed, which changes at unknown transition times and where each segment is modelled as a polynomial with known order but unknown parameters. A model order method based on the maximum likelihood principle is suggested, by imposing the constraint that the complete signal is continuous, for jointly determining the number of segments, the transition times and the parameters of each polynomial. Simulation results show that the proposed approach outperforms the unconstrained segmentation.

Offline segmentation and online language processing units

Although metalinguistic (‘offline’) awareness of words as linguistic units has been related to literacy, it is still uncertain whether literacy also affects the units of language that people process unconsciously (‘online’). In this contribution, we first discuss the characteristics of offline and online tasks, opening up the perspective that such tasks vary in nature along a continuum ranging from more offline to more online. Subsequently, we present a study employing three relatively more offline and two more online tasks which we conducted among 83 preliterate and 121 literate children at Dutch primary schools. The results of the more offline tasks reveal a significant influence of literacy on segmentation along word-boundaries, while the results from the relatively more online tasks are less clear-cut with respect to the way in which literacy affects language processing. Keywords: literacy; language acquisition; metalinguistic awareness; online and offline task; word-segmentation; language processing; developmental psycholinguistics; Dutch

Bayesian Change-Points Estimation Applied to GPS Signal Tracking

A hierarchical Bayesian model is applied to off-line segmentation of the GPS signal discriminator. The purpose of this work is to estimate the code delay of the receiving GPS CDMA code in order to retime the local receiver code and to estimate the pseudorange satellite receiver. The goal of our approach is to obtain a high-rate accurate positioning in the dynamic navigation case. We show that the behaviour of the coherent discriminator of a GPS pilot channel can be modelized by a piecewise stationary process. In our approach the discriminator behaviour in each stationary segment is approximated by a constant acceleration model, and the code delay at each end of the segments is known. The interest of this approach is that we use the coherent values of the discriminator in each segment to estimate the change instants of the process and to get in this case an accurate estimation of the code delays. In this context, a simultaneous estimation of the change instants is considered. We define the a posteriori distribution which integrates in its expression the signal change instants and the parameters of its statistical model. The proposed model leads after marginalization to a penalized contrast function that we minimize to estimate the discriminator change instants. The interest of the proposed model is that we can integrate in our estimate prior information on the roughly known values of the signal-to-noise ratio and relative speed satellite receiver. The potential of the proposed method is shown on experimentations realized on synthetic and real data for millisecond receiver localization.