Search

Human attention tends to get focused on the most prominent components of a scene which are in sharp contrast with the background. These are termed as salient regions. The human brain perceives an object to be salient based on various features like the relative intensity, spread of the region, color contrast with the background, size and position within an image. Since these features vary widely, no crisp thresholds can be specified for an automatic salient region detector. In this paper we present a rule based system which uses a set of fuzzy features to mark out the salient region in an image. A genetic algorithm based evolutionary system is used to learn the rules from the training images. Extensive comparisons with the state-of-the-art methods in terms of precision, recall and F-measure are made on two different publicly available datasets to prove the effectiveness of this approach. The application of the proposed salient object detection approach is shown in non-photorealistic rendering, perception based image compression and context aware retargeting applications with promising results.

In this paper we explore detection of unusual activities using Markov Logic Network (MLN) based approach. Any human activity which is in variance from a defined usual set attracts human attention and is considered unusual. Such activities include anomaly detection in crowds, some repetition or omission of subactivities in a given sequence of activities in Ambient Assisted Living environments or an outlier in case of surveillance. In this paper, we target the unusual activities occurring in workplaces. Typical usual activities considered are: entering a room, walking, sitting down and working. We define activities with unlabeled actions as well as outliers as unusual activities. The outliers include activities with repeated actions and activities with certain actions omitted. We use Markov Logic Network because it allows us to create common sense rules defining the relationship between different actions and activities. We validate our results on a noisy data set.

Human attention tends to get focused on the most prominent components of a scene which are in sharp contrast with the background. These are termed as salient regions. Saliency is defined in terms of local and global feature contrasts. The human brain perceives an object of salient type based on its difference with the surroundings in terms of color and texture. There have been many color based approaches in the past for salient object detection. In this paper, we define the uncertainty of a window being salient or background in terms of information extracted from different color components. The uncertainty associated with the elements of a fuzzy set is described by a membership function, which gives the degree of association of each element to the set. The overall uncertainty is sought to be quantified by an entropy function. To locate the salient parts of the image, we make use of the entropy to compute a new set of features from color and luminance components of the image. Extensive comparisons with the state-of-the-art methods in terms of precision, recall and F-Measure are made on a publicly available dataset to prove the effectiveness of this approach.

In this paper we propose to summarize videos based on key frames. We improve upon the histogram and pixel difference based approach with fuzzy rule based approach and also give a new approach which reduces the computation of framewise differences. We test our methods using fidelity ratio and compression ratio on videos of sports from YouTube and UCF sports dataset, videos of commercials and sitcoms. The results of our methods are seen to be comparable to other state of the art approaches.

Color is the most dominant feature used by the human brain to perceive an image region to be salient. Since size, shape, location and colors of salient objects vary widely, a fuzzy rule based system is proposed in this paper which uses color proximity, color spread, connected components and presence of a face as linguistic variables. These rules are learned using a genetic algorithm. The colorspace used is the CIELab colorspace which closely conforms with human perception of colors. A publicly available image dataset is used for training and testing the system. Comparisons with existing state-of-the-art methods in terms of precision, recall and F-Measure have been presented.

Automatic salient region detection in an image is useful for image compression, image cropping for resizing on smaller displays, object recognition and tracking. In this paper case-based reasoning is used to make a system learn the relevant attributes of a salient region, based on global and local color contrast. CIELab colorspace is used as it is perceptually uniform and matches the human visual perception. The parameters used are background colors at the boundary, color distance, spatial variance, size of the connected components of salient color and dominant colors. Intensity values are used to deal with images containing black and white shades. Exemplar images are presented to the system to categorize the cases. The method is tested on a large image database.