A Bayesian network-based tunable image segmentation algorithm for object recognition

Abstract

We present a Bayesian network-based tunable image segmentation algorithm that can be used to segment a particular object of interest (OOI). In tasks such as object recognition, semantically accurate segmentation of the OOI is a critical step. Due to the OOI consisting of different-looking fragments, traditional image segmentation algorithms that are based on the identification of homogeneous regions tend to oversegment. The algorithm presented in this paper uses Multiple Instance Learning to learn prototypical representations of each fragment of the OOI and a Bayesian network to learn the spatial relationships that exist among those fragments. The Bayesian network, as a probabilistic graphical model, in turn becomes evidence that is used for the process of semantically accurate segmentation of future instances of the OOI. The key contribution of this paper is the inclusion of domain-specific information in terms of spatial relationships as an input to a conventional Bayesian network structure learning algorithm. Preliminary results indicate that the proposed method improves segmentation performance.

The emerging era of genomic data integration for analyzing splice isoform function

“…An advantage of this approach is that it covers all isoforms that are captured in the RNA-seq datasets, compared to structure-based methods that only cover those that have clear domain information. Furthermore, any classification method can be built into this MIL framework, such as support vector machines [82], Bayesian networks [83], and K-nearest neighbors [84]. This approach was recently used to generate genome-wide predictions of isoform functions and identified functional differences for a large number of multi-isoform mouse genes [16].…”

Incorporating shape into spatially-aware adaptive object segmentation algorithm

“…A comparison with ground-truth segmentations between segmentation with shape descriptions and segmentation without shape descriptions is made in Table 3 with the Q value for each segmentation is shown. The table compares results from our new algorithm (visual+shape and spatial+shape) with the results from our previous algorithm in [1](visual and visual+spatial). Results for most of the test images show significant improvement in incorporating shape descriptions in the segmentation process and agree more with the groundtruth segmentations.…”

“…Our recent work [1] presented a tunable object-specific segmentation algorithm that uses Bayesian Networks (BN) as a probabilistic graphical model to learn both visual and spatial relationships among the components of the OOI. The performance of this algorithm heavily depends on the assumption that the visual appearance variability among OOI training instances is low.…”

“…In this paper, we present an extension to this algorithm that relaxes this assumption by incorporating shape information into the OOI model. If the assumption of low visual appearance variability among OOI instances does not hold, the segmentation performance of the algorithm in [1] is low. This is because common instances among OOI regions are not identified due to their high visual variability.…”