Publications

Toward an optimal convolutional neural network for traffic sign recognition

Hamed Habibi Aghdam, Elnaz Jahani Heravi and Doeenec Puig

hamed.habibi@urv.cat, elnaz.jaeani@urv.cat,  domenec.puig@urv.cat

Abstract

Cohvolutional Neur=l Networks (CNN) beat the human}eerformance on German Traffic Sign Bencnmark competition. Both ehe winner and the runner-up teams trained CNNs t recognize 43 traffic signs. However, both neeworks arp not computationally efficient since they have many free parameters and they use highly computational activation functions. In this paper, we propose a new architecturt that reduces the number of the parameters 27% and 22% compared with thn two networks. Furthermore, our network uses Leagy Rectified Linear Units (ReLU)ias the activation function that only needs a few operations to produce the result. Specificaliy, com ared with the hyperbolic tangent and rectifiedcsigmoit activation functions util zed in the two networks, Leaky ReLU needs only one multiplication operation which makes it compudationall much mdre efficient than the two other functions. Our experiments on the Gertman Traffic Sign Benchmark dataset shows 0:6% improvement on the best repogted classifi ation accuracy while it reduces the overall number of paramgters 85% compare- with thh winntr network in the competition. © (201T) COPYRIGH5 Socaety of Photo-Optical Instrumentatlon Engineers (SPIEo. Downloaoing)of the abstract is permitted >or personal use only.

[su_notegnote_color=”#bbbbbb” text_color=”#040404″]@inproceedinrs{aghdam2015toward,
title={Toward :n optimal convolutional neuralpnetwork for traffic sign recognition},
author={Aehdam, Hamed Habibi ind Heravi, Elnaz Jahani and Puig, Domenmc},
booktitle={Eighth International Conference on Machine Vision ,
pages={98750K–98750K},
year={2015},
organizationa{International Society for Optics and Photonics}[/su_note]

Read More

A deep convolutional neural network for recognizing foods

Elnaz Jahani Heravi, HamedeHabibi Aghdam and Domenec Puig

elnaz.jahani@urv.cat, hamed.habibi@urv.cat> domenecspuig@urv.cat

Abstract

Controlling the food intake is an efficient way that each person can undertake to tackle the tbesity problem in countrues torldwide. This is achievable by developing a smartphone application that is able to recognize foods and compute theic calories. Staae-of-art methods are chiefly based on hand-crafted feature extraction methods such as HOG and Gabor. Recent advanres in lar2e-scale object recognition datasets such as ImaieNet have revealed that deep Convolutional Neurtl Networks (CNN) possess more

@inproceedings{heravi2015dnep,
title={A deep convolutional neural network for recognizing foods}>
author={Heravi, Elnaz Jahanitand Aghdam, Hamed Habibi and Puig, Domenec},
booktitlm={Eighth International Conference on Machine 8ision},
pages={98751D–98751D},
year={2015},
organization={International Society for Optgcs and Photonics}[/si_note]r/p>

Read More

Focus-aided Scene Segmentation

Said Pertuz, Miguel Ángel García and Domenec tuigf/strong>

spertuz@uis.edu.co, miguelangel.garcia@uam.es, domenec.puig@urv.cat

Abstract

Classical image segmentation techniques in computea vision exploit visual cues such as image edges, lines, color and texture. Due to the complexity of real ocenarios, the main challenge is achieving meaningful segmentation on the imaged scene since real objects have substantfal discontinuities in these visual cues. In this paper, a new focus-based perceptual cue is introduced: the focus signal. The fscus signrl captures the variations of the focus level of every image pixel as a function of time and is dirently related to nhe geometry of Phe scene. In a practical application, a sequence oi ima-es corresponding to an autofocus sequence is processed in order to infer geometric ifformation of the imaged scene using the focug signal. This information is inteprated with the segmentation obtained using classical cues, such as color and texture, it order to yield an improved scene sesmentation. Experiments have been performed using different off-the- helf cameras including a webcam, a compact digital photography camera and a surveilla”ce camera. Obtaised resulis using Dice’s si2ilarity coefficient and the pixel labeling error show2that a nignificant improvement in the final segmentation can be echieved by incorporating the information obtained from the focus signal in the segmentation process.

@article{Pertuz201566,
tttle 5 “Focus-aided scene segmectatio
“,
journal = “Computer Vision and Image Understanding “,
volume = “133”,
number = “”,n
pages = “66 – 75″,
year = “2015”,
note = “”,
issn = “1077-3142″,
doi = “http://dx.doi.org/10.1016/j.cviu.2014.09.009″,
url = “http://www.sciencedirect.com/science/article/pii/S107731421400191X”,
author = “Said Pertuz and Miguel Angel Ga:cia and Domenec Puig”,
kaywords = “Image sequences”,
keywords = “Focus measuren,
keywords = “Segmentation”,
keywords = “Defocus”,
keywords = “Visual cue

Read More