25-08-2014, 12:04 PM
Computational Perceptual Features for Texture
Representation and Retrieval
Computational.pdf (Size: 1.76 MB / Downloads: 25)
Abstract
A perception-based approach to content-based image
representation and retrieval is proposed in this paper. We consider
textured images and propose to model their textural content by
a set of features having a perceptual meaning and their application to content-based image retrieval.
I. INTRODUCTION
TEXTURE has been extensively studied and used in liteature since it plays a very important role in human visual
perception.. Although there exists no precise and universal definition of texture, some intuitive concepts can be defined about
texture. Texture refers to the spatial distribution of grey-levels
and can be defined as the deterministic or random repetition
of one or several primitives in an image.Microtextures refer
to textures with small primitives while macrotextures refer to
textures with large primitives [31]–[33].
II. RELATED WORKS
There are some works published in literature on the subject of
human visual perception since the early studies done by Julesz
[21] and Bergen et al. [10]. However, there are two main works
that are closely related to our work. The first work is done by
Tamura et al. [30] and the second work is done by Amadasun et
al. [8]. Each of the two has proposed computational measures
for a set of textural features. The work of Tamura et al. [30] was
based on the co-occurrence matrix and the work of Amadasun et
al. [8] was based on a variant of the co-occurrence matrix called
NGTDM (neighborhood grey-tone difference matrix). The objective that we follow in our work falls into this global
framework.. We propose, however, a new method to estimate a
set of perceptual textural features.
The rest of this paper is organized as follows. we present definitions of the set of perceptual textural features
we are considering in this study; in Section IV, the autocor
PERCEPTUAL TEXTURAL FEATURES
However, only a small list of features is considered as
the most important.. This list comprises coarseness, contrast and
directionality.. Other features of less importance are busyness,
complexity, roughness and line-likeness [8], [30]. In this study . In the following, we
give conceptual definitions of each of these features
Coarseness is the most important feature and, in a certain
sense, it is coarseness that determines the existence of texture
in an image.
Directionality is a global property in an image. It measures
the degree of visible dominant orientation in an image.
Contrast measures the degree of clarity with which one can
distinguish between different primitives in a texture.
V. COMPUTATIONAL MEASURES FOR TEXTURAL FEATURES
In order to simplify the presentation As explained earlier, we are using two representations
or viewpoints: the computational measures presented in the
following are computed on both the original images and the
autocorrelation function
A. Coarseness Estimation
When we consider the autocorrelation function, one can
notice two phenomena related to coarseness : 1. Coarseness is
saved in the corresponding autocorrelation function; 2. For fine
textures, the autocorrelation function presents a lot of local
B. Contrast Estimation
When considering the autocorrelation function, we can notice
that the value of this function decreases quickly
C. Directionality Estimation
Regarding directionality, we want to estimate two parameters:
the dominant orientation(s) and the degree of directionality
PSYCHOMETRIC METHOD
• Considering the consolidated human rankings of textures
according to each textural feature and the computational
rankings obtained for each textural feature,we compute the
rank-correlation between the two rankings for each textural
feature.
A. Sum of Rank Values
Suppose that images were ranked in different rankings,
each performed by a human subject
B. Spearman Coefficient of Rank-Correlation
Once the different consolidated human rankings are obtained
for each textural feature,and considering the computational
ranking for the same textural feature,we compute the rank -correlation between the two rankings.
I. EXPERIMENTAL RESULTS AND
PSYCHOMETRIC EVALUATION
Psychological experimentations were conducted with human
subjects in order to evaluate the correspondence between
computational results obtained by applying the proposed computational measures and those obtained with human subjects.
A. Computational and Consolidated Human Rankings
Table I summarizes the computational rankings for each of
the four textural features. Table II summarizes the consolidated
human rankings for each of the four textural features.
B. Correspondence Between Human and Computational
Rankings
Table III gives the Spearman coefficient of rank-correlation
between the two rankings, consolidated human ranking and
computational ranking,computed using (17) and taking into
account the case where several images are given the same rank
C. Features Relatedness
Table IV summarizes rank-correlation between consolidated
human rankings and Table V summarizes rank-correlation
between computational rankings for each of the four textural
features
D.Comparison With Related Works
A comparison between the results we obtained and the results
obtained by the two main related works, namely Tamura et al.
[30] and Amadasun et al.
VIII. APPLICATION TO CONTENT-BASED IMAGE RETRIEVAL
Content-based image retrieval is the ability to search an image
d databases and retrieve relevant results using the content of images.
A. Similarity Measure
The similarity measure used is based on the Gower coefficient
of similarity [15] we have developed in our earlier work [4].
B. Results Fusion
In order to merge the results returned by each of the two
representations, we have experimented several results merging
models.
C. Brodatz Database
We have applied the computational features presented in this
paper in a large image retrieval experience on Brodatz database
E. Precision and Recall Measures
Precision and recall measures are widely accepted and used
to benchmark search relevance (effectiveness) in information retrieval systems. Precision, which can be defined as the number
of relevant and retrieved images divided by the number of retrieved images, measures the ability of a model to reject nonreevant images with respect to a query.
F. Comparison to Related Works
Comparing one’s results to other works is not an easy task. In
fact, in order to do so, all works must use the same database, the
same queries and the same evaluation criteria. In practice, these
are rarely available.
IX. CONCLUSION
A new perceptual model based on a set of computational
measures corresponding to perceptual textural features, namely
coarseness, directionality, contrast, and busyness, was introduced in this paper. Computational measures are based on two
different representations (viewpoints): original images and the
autocorrelation function associated with images. Coarseness
was estimated as an average of the number of extrema. Contrast
was estimated as a combination of the average amplitude of the
gradient, the percentage of pixels having the amplitude superior
to a certain threshold and coarseness itself. Directionality was
estimated as the average number of pixels having the dominant
orientation(s). Busyness was estimated based on coarseness