MACHINE VISION AND IMAGE PROCESSING
Year:2017 | Volume:4 | Issue:1 |Papers Count:6

Object recognition in cluttered background is a challenging problem in computational modeling. When objects are present on natural backgrounds, the performance of object recognition models drop significantly. However, humans recognize objects accurately and swiftly despite this challenging condition. It seems that, our visual system achieves this ability based on lateral connections and feedback connections from higher areas.One of the computational object recognition models that recently has achieved a remarkable performance in object recognition is convolutional neural network (CNN). It resembles feed-forward sweep of visual information processing. In this study, based on CNNs and inspired by biological evidence we proposed a recurrent object recognition model. The model simulates recurrent dynamics of visual object processing by implementing feedback and lateral connections. Evaluating the model to recognize objects on natural background, we showed that the proposed mechanisms significantly improves performance. In addition, visualizing the representations of layers indicatedthat deeper layers of the CNNs remove the background much better than the lower layers. According to the results, using both mechanisms -the feedback from higher layers and the interlayer surround suppression mechanisms- simultaneously in structure of CNN, the performance improvement was more than when either one was usedalone.This observation is in accordance withthe biological evidence from the human visual system.

In the image stegenography, the distortion of stego image will be increased by increasing the number of embedded bits and this makes the enemy be much more suspicious about the existence of secret information in the images. One can use the matrix embedding technique to alleviate this problem which can be done by dividing the cover image into some blocks and limiting the changes in each block. However, the embedding capacity will be limited by this technique. In this article, a solution to this problem (limited embedding capacity), a modified version of Fan’s method, is presented. In this technique, the maximum number of changes in each block is chosen to be 2 instead of 1 in Fan’s technique which leads to reducing the blocks length and increasing the embedding capacity. The results revealed that for a given capacity, the stego image distortion is decreased and the embedding capacity is increased in compared with the other existing methods in this field. Also, the stego images have high PSNR along with high vision quality which leads to more security.

After vehicle detection and vehicle type recognition, it is vehicle make and model recognition (VMMR) that has attracted researchers attention in the last decade. This problem is known as a hard classification problem due to the large number of classes and small inner-class distance. This paper is proposed a new method for recognition of make and model of vehicles.The proposed approach has two parts. A new part-based approach for vehicle make and model recognition and a new method for auto extraction of parts. This approach concentrates on meaningful parts of vehicle like lights, grilles and logo for classification of different classes. The Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM) have been used for feature extraction and classification tasks respectively. For evaluation purposes, a dataset including 720 images from frontal and rear view of21 different classes of vehicles have been prepared and fully annotated based on their parts. The experimental results showed the effectiveness of the part-based approach in compare to the traditional approaches and the high accuracy gained from auto extracted parts.The proposed method achieved 100% accuracy on both frontal and rear view.

Word synthesis and generating Persian handwriting documents is an interesting topic and can help to improving the results in similar fields such as writer identification and Optical Character Recognition. In this paper we propose a new method to create automatic handwriting words based on previous author’s writing. For this propose at first some control points are defined on Persian characters. After that interpolation is used for connecting isolated characters. Different interpolationfunctions are used to creating natural handwriting based on different forms of characters connection and their locations.Mean opinion scores show that our synthesis words have about 80% similarity with writer handwriting documents.

The principal component analysis (PCA) is one of the procedures that have been a successful performance in signal processing and dimension reduction of the signals. However, a requirement in applying PCA to the images is converting images into a vector.This process leads to loss spatial locality information. To solve this problem, the two-dimensional PCA was proposed. Also, most recently the sparse principal component was introduced that not only keep the properties of standard PCA but also try to make a lot of elements of the basis vectors to zero. In this paper, inspired by the above two ideas, the two-dimensional sparse principal component analysis (2-D. SPCA) is proposed.In this paper, the Least Angle Regression- Elastic Net formula, in addition, using l1 and l2 constraints is extended to two-dimensional model with a few minor changes in its input to approach 2-D. SPCA.The two-dimensional sparse principal component analysis is evaluated in image compression. Before applying the algorithm, the image is divided into several blocks with resolution 8×8 and a database of these blocks is formed. Comparison the performance of 2-D. SPCA and Discrete Cosine transform, for the same number of elements that are necessary to save the image after the conversion shows the good performance of the proposed algorithm. In addition, the proposed algorithm is applied to 8×8 blocks of 60 images with different textures, and the resulted two-dimensional sparse principal components could be used for other test images with a suitable performance.

With advance of technology, manipulation and distribution of digital media such as image, audio and video has been widespread explosively. This causes problems like unauthorized copy being more important. Watermarking is a method for copyright protection and hidden information transfer. Wavelet transform has good functionality in various watermarking methods because of its good ability of modeling human visual system. Wavelet transform extensions like basic wavelet are utilized in variety applications of image processing and good results have been reported from them. No comprehensive research has been done in efficiency of extended wavelets on watermarking robustness. In this paper efficiency of wavelet transform extensions is investigated in different attacks. For this purpose a general framework is proposed for embedding and watermark extraction based on bidiagonal singular value factorization. Results show classic wavelets have good robustness against attacks like compression and noises, but they have less robustness against geometric attacks such as rotation and shearing. Unlike classic wavelets, multidirectional wavelet extensions are robust in geometrical attacks.