V-SIN: Visual Saliency detection in noisy Images using convolutional neural Network

In the Computer era, the capability of a machine to differentiate salient object from the background has became a critical fact in the domain of Computer Vision. The features in the noisy images get greatly compromised, owing to which the Salient Object Detection (SOD) is difficult. Moreover, the existing research has not yet been matched the performance of humans for detecting the visual saliency in noisy environment. Therefore, this work highlights a novel SOD technique in noisy environment using convolutional neural network (CNN), while the salient object detection accuracy has been well maintained. The denoising of the image is performed using CNN which comprises the coordinate descent as a regularizing function. The performance of our proposed V-SIN technique has been assessed with four evaluation parameters, computing time, recall, precision, and F-measure on two publicly available image datasets. The experimental evaluations on these two dataset shows that the proposed model has been much robust to detect salient object in the presence of noise or mixture of noises in images.