SDSP: A novel saliency detection method by combining simple priors

Zhang, Lin; Gu, Zhidong; Li, Hongyu

doi:10.1109/icip.2013.6738036

Cited by 151 publications

(86 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…An image's visual saliency map can be used as a weighting function to indicate the importance of a local region to the human visual system [26]. In our work, we adopt the idea in [27] to obtain a visual saliency map, which has been proved to be effective, simple, and can generate excellent results in most cases. An image from the STL-10 dataset and its corresponding visual saliency map are illustrated in Figure 1.…”

Section: Constructing Local Patch Vectorsmentioning

confidence: 99%

Local Patch Vectors Encoded by Fisher Vectors for Image Classification

et al. 2018

View full text Add to dashboard Cite

Abstract:The objective of this work is image classification, whose purpose is to group images into corresponding semantic categories. Four contributions are made as follows: (i) For computational simplicity and efficiency, we directly adopt raw image patch vectors as local descriptors encoded by Fisher vector (FV) subsequently; (ii) For obtaining representative local features within the FV encoding framework, we compare and analyze three typical sampling strategies: random sampling, saliency-based sampling and dense sampling; (iii) In order to embed both global and local spatial information into local features, we construct an improved spatial geometry structure which shows good performance; (iv) For reducing the storage and CPU costs of high dimensional vectors, we adopt a new feature selection method based on supervised mutual information (MI), which chooses features by an importance sorting algorithm. We report experimental results on dataset STL-10. It shows very promising performance with this simple and efficient framework compared to conventional methods.

show abstract

Section: Constructing Local Patch Vectorsmentioning

confidence: 99%

Local Patch Vectors Encoded by Fisher Vectors for Image Classification

et al. 2018

View full text Add to dashboard Cite

show abstract

“…However, no one tries to use VS-based pooling strategy in the objective quality assessment of tonemapped images. In this paper, we tried 5 kinds of VS models, but in this section we only give a brief introduction about two of them which we found have outstanding performance when used in IQA [5] or saliency detection [7], namely SR and SDSP.…”

Section: Two Visual Saliency Modelsmentioning

confidence: 99%

“…Saliency detection by combining simple priors (SDSP) was proposed in [7]. This method was inspired by three simple priors.…”

Section: Sdsp: Saliency Detection By Combining Simple Priorsmentioning

confidence: 99%

Integrating Visual Saliency Information into Objective Quality Assessment of Tone-Mapped Images

Liu

Zhang

et al. 2014

Intelligent Computing Theory

Self Cite

View full text Add to dashboard Cite

Abstract. Tone-mapped images are the low dynamic range (LDR) images converted from high dynamic range (HDR) images. Recently, the objective quality assessment of tone-mapped images is becoming a challenging problem. However, there is no mature algorithm to deal with this issue until the tone-mapped image quality index (TMQI) was proposed recently, which is tone-mapped image quality index (TMQI). Unfortunately, the pooling method of the structural fidelity map in TMQI is the simple "mean", which makes the result unsatisfying. On the other hand, recent studies have found that different locations of an image may have different contributions to the quality perception of the human visual system. The significance of a local image region can be well characterized by a visual saliency (VS) model. Inspired by this insight, in this paper, we propose a VS-based pooling strategy for the objective quality assessment of tone-mapped images. The experimental results clearly demonstrate the efficacy of our proposed method.Keywords: Tone-mapped images, high dynamic range images, objective quality assessment, visual saliency. IntroductionRecently, researchers have shown a growing interest in high dynamic range (HDR) images. Compared to the low dynamic range (LDR) images, the range of intensity levels of HDR images is largely wider, thus enjoying advantages. Specifically, the range of intensity level of HDR images could be on the order of 10000 to 1, which allows for accurate representations of luminance variations in real scenes, ranging from direct sunlight to faint starlight [1], while the LDR images only have 256 intensity levels. However, General displays are designed for displaying LDR images, so we must use tone-mapping operators (TMOs) to create the corresponding LDR images for the visualization of HDR images. As the reduction in dynamic range, the LDR images converted from HDR images cannot preserve all the information. With an increasing number of TMOs being developed, a question appears, that is which TMO has a better performance? This question drives us to develop an index to compare them, and then utilize the index for optimizing parameters in TMOs.

show abstract

“…Some models such as [13], consider center sensitivity to generate saliency maps. In this work, a 2D anisotropic Gaussian intensity map is used to generate a center sensitivity map.…”

Section: Proposed Algorithmmentioning

confidence: 99%

“…A Discrete Cosine Transform (DCT) based saliency model is proposed in [12]. A fast saliency detection algorithm proposed in [13] considers centre sensitivity, however, achieves low prediction accuracy. The authors of [14] sample the image into rectangular regions of interest and then compute the local saliencies.…”

Section: Introductionmentioning

confidence: 99%