Saliency histogram equalisation and its application to image resizing

Hsin, Hsi-Chin

doi:10.1049/iet-ipr.2015.0559

Cited by 9 publications

(45 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Pixels with great saliency values are often in the ROI, which need to be preserved as much as possible; on the other hand, those with small saliency values in the region of non‐interest (RON) can be ignored in the retargeted images/videos. This section reviews the SC algorithm [2] that is representative of the discrete approach, the SHE algorithm [13] that is classified into the continuous approach, and the improved SHE algorithm by combining SHE with SC [18].…”

Section: Related Workmentioning

confidence: 99%

“…In the continuous approach, both of the original and the retargeted images are represented by mesh grids. Based on the SH of an image, which shows the occurrence of salient objects, the SHE algorithm had been proposed to construct a non‐uniform mesh for the original image [13]. Specifically, the marginal SH (MSH) obtained by

p_{y} false(y false) = \frac{\sum_{x = 1}^{M} p (x, y)}{\sum_{x = 1}^{M} {false\sum}_{y = 1}^{N} p false(x, y false)}; y \in [1, N]

is used to partition an

M \times N

image, I , into L vertical strips with an equal accumulated amount of saliency values as follows:

v_{n} = \underset{k}{arg max} \frac{n - 1}{L} < {false\sum}_{y = 1}^{k} p_{y} false(y false) \leq \frac{n}{L}

where

p false(x, y false)

is the SH of I , and

v_{n}

is the coordinate of the lower‐right vertex of the n th vertical strip in the y ‐direction,

n = 1, \dots, L

, with the boundary condition

v_{L} = N

.…”

Section: Related Workmentioning

confidence: 99%

“…As shown in Fig. 4, the first step is to construct the upper and lower meshes

M_{k}^{u}

and

M_{k}^{l}

for

F_{k}^{u}

and

F_{k}^{l}

using the SHE algorithm [13]. The second step is to divide the combination of

M_{k}^{u}

and

M_{k}^{l}

into the left and right portions with respect to the GCP,

v_{c, k}

, which are used as the initial meshes of SHE to construct the left and right meshes

M_{k}^{f}

and

M_{k}^{r}

for

F_{k}^{f}

and

F_{k}^{r}

, respectively.…”

Section: Video Retargeting In the Saliency Histogram Domainmentioning

confidence: 99%

“…Conger et al proposed multi‐SC [5] with wavelet filters [6]. In [7], seams of wavelet trees were constructed for image resizing at the scale level in the wavelet domain.…”

Section: Introductionmentioning

confidence: 99%

“…As shown in the study of human visual system, the saliency map (SM) of an image is crucial to the development of modern image retargeting [10–12]. In [13], the normalised SM called the saliency histogram (SH) was used as a probability density function to provide the presence of prominent objects, and the SH equalisation (SHE) algorithm was proposed to construct non‐uniform meshes for effective resizing of images.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Video retargeting based on SH equalisation and seam carving

Hsin

2019

IET Image Processing

Self Cite

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 99%

p_{y} false(y false) = \frac{\sum_{x = 1}^{M} p (x, y)}{\sum_{x = 1}^{M} {false\sum}_{y = 1}^{N} p false(x, y false)}; y \in [1, N]

is used to partition an

M \times N

image, I , into L vertical strips with an equal accumulated amount of saliency values as follows:

v_{n} = \underset{k}{arg max} \frac{n - 1}{L} < {false\sum}_{y = 1}^{k} p_{y} false(y false) \leq \frac{n}{L}

where

p false(x, y false)

is the SH of I , and

v_{n}

is the coordinate of the lower‐right vertex of the n th vertical strip in the y ‐direction,

n = 1, \dots, L

, with the boundary condition

v_{L} = N

.…”

Section: Related Workmentioning

confidence: 99%

“…As shown in Fig. 4, the first step is to construct the upper and lower meshes

M_{k}^{u}

and

M_{k}^{l}

for

F_{k}^{u}

and

F_{k}^{l}

using the SHE algorithm [13]. The second step is to divide the combination of

M_{k}^{u}

and

M_{k}^{l}

into the left and right portions with respect to the GCP,

v_{c, k}

, which are used as the initial meshes of SHE to construct the left and right meshes

M_{k}^{f}

and

M_{k}^{r}

for

F_{k}^{f}

and

F_{k}^{r}

, respectively.…”

Section: Video Retargeting In the Saliency Histogram Domainmentioning

confidence: 99%

“…Conger et al proposed multi‐SC [5] with wavelet filters [6]. In [7], seams of wavelet trees were constructed for image resizing at the scale level in the wavelet domain.…”

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations