International Journal of Image and Graphics

Dense trajectories have been shown as a very promising method in the human action recognition field. In this paper, we propose a new kind of video descriptor, generated from the relationship between the trajectory’s optical flow with the gradient field in its neighborhood. Orientation tensors are used to accumulate relevant information over the video, representing the tendency of direction in the descriptor space for that kind of movement. Furthermore, a method to cluster trajectories using their shape is proposed. This method allows us to accumulate different motion patterns in different tensors and easier distinguish trajectories that are created by real movements from the trajectories created by the camera’s movement. The proposed method is capable to achieve the best known recognition rates for methods based on the self-descriptor constraint in popular datasets — Hollywood2 (up to 46%) and KTH (up to 94%).

Image compression is the emerging field to transmit the multimedia products like image, video and audio. Image compression is used to reduce the storage quantity as much as possible. The objective of this paper is to compare the multiscale transform based image compression encoding techniques. The multiscale transforms involved in this paper are wavelet transform, bandelet transform, curvelet transform, ridgelet transform and contourlet transform. Wavelet transform allows good localization both in time and frequency domain. Bandelet transform takes geometric regularity of the natural images to improve the efficiency of representation. Curvelet transform handles curve discontinuities well. Curvelets are the good tool for the analysis and the computation of partial differential equations. Curvelets also have micro local features which make them especially adapted to certain reconstruction problems with missing data. The ridgelet transform is the core idea behind curvelet transform. It is used to represent objects with line singularities. The contourlet transform gets smooth contours and edges at any orientation. It filters noise as well. The Encoding techniques involved in this paper are spatial orientation tree wavelet (STW), set partitioned embedded block (SPECK) and compression with reversible embedded wavelet (CREW). The performance parameters such as peak signal to noise ratio (PSNR), image quality index and structural similarity index (SSIM) are used for the purpose of evaluation. It is found that bandelet transform with all the encoding techniques work well.

One of the most effective ways of expressing emotion is through facial expressions. This work proposes and discusses a geometrical descriptor based on the calculation of distances from coordinates of facial fiducial points, which are used as features for training support vector machines (SVM) to classify emotions. Three data sets are studied and six basic emotions are considered in our experiments. In comparison to other approaches available in the literature, the results obtained with our geometrical descriptor demonstrated to be very competitive, achieving high classification F-score rates. Additionally, we evaluate whether the combination of our geometrical descriptor with an appearance feature, the Gabor filter, allows emotions to be even more distinguishable for the classifier. The result is positive for two out of three data sets. Finally, to simulate in-the-wild scenarios, an active appearance model (AAM) is trained to position the fiducial points on the correct facial locations, instead of using the ones provided by the data sets. As the fitting error is considered acceptable, the former experiments are also conducted with the new data generated by the AAM. The results show a small drop on the F-score values when compared to the data originally provided by the data sets,but are still satisfactory.

Copula were introduced in Markov models in the early 2000s to better model the relationship between the observation data involved in these models. However, their estimation is difficult. This paper presents a new approach in the estimation of copula in Markov models. The proposed approach is based on a nonparametric method of estimating the density of the copula. The decomposition of an orthonormal basis of the unit interval support polynomials is used to estimate this density. The family of polynomial used is built from the family of Legendre polynomials. Our approach has the major advantage of reducing the problem of unsupervised image segmentation by Pairwise Markov chains to the problem of the estimated marginal distributions unlike the conventional approach which requires both an estimation of the density the marginal distributions and the estimation of the density distribution of the copula. Moreover, the problems of boundary effect encountered in the density estimation of copulas are solved by the use of orthonormal basis functions in the unit interval. The new model is validated from experiments performed on synthetic images and real images including optical and radar satellite images not necessarily affected by the Gaussian noise. The results are encouraging and show the proposed model as an interesting alternative to Pairwise Markov chains commonly used in literature.

Extraction of global face appearance and local interior differences is essential for any face recognition application. This paper presents a novel framework for face recognition by combining two effective descriptors namely, Zernike moments (ZM) and histogram of oriented gradients (HOG). ZMs are global descriptors that are invariant to image rotation, noise and scale. HOGs capture local details and are robust to illumination changes. Fusion of these two descriptors combines the merits of both local and global approaches and is effective against diverse variations present in face images. Further, as the processing time of HOG features is high owing to its large dimensionality, so, the study proposes to improve its performance by selecting only most discriminative HOG features (named discriminative HOG (DHOG)) for performing recognition. Efficacy of the proposed methods (DHOG, $\mathrm{\text{ZM}}+\mathrm{\text{HOG}}$ and $\mathrm{\text{ZM}}+\mathrm{\text{DHOG}}$) is tested on ORL, Yale and FERET databases. DHOG provides an improvement of 3% to 5% over the existing HOG approach. Recognition results achieved by $\mathrm{\text{ZM}}+\mathrm{\text{HOG}}$ and $\mathrm{\text{ZM}}+\mathrm{\text{DHOG}}$ are up to 15% and 18% higher respectively than those obtained with these descriptors individually. Performance is also analyzed on LFW face database and compared with recent and state-of-the-art methods.

Medical image fusion is being used at large by clinical professionals for improved diagnosis and treatment of diseases. The main aim of image fusion process is to combine complete information from all input images into a single fused image. Therefore, a novel fusion rule is proposed for fusing medical images based on Daubechies complex wavelet transform (DCxWT). Input images are first decomposed using DCxWT. The complex coefficients so obtained are then fused using normalized correlation based fusion rule. Finally, the fused image is obtained by inverse DCxWT with all combined complex coefficients. The performance of the proposed method has been evaluated and compared both visually and objectively with DCxWT based fusion methods using state-of art fusion rules as well as with existing fusion techniques. Experimental results and comparative study demonstrate that the proposed fusion technique generates better results than existing fusion rules as well as with other fusion techniques.