MACHINE VISION AND IMAGE PROCESSING
Year:2016 | Volume:2 | Issue:2 |Papers Count:6

Mineral segmentation in thin sections based on image processing algorithms is one of the popular research topics geosciences. Rocks are the main information resource for geological studies, and mounting thin section from rocks is the most popular method for studying them. Mineral segmentation in thin sections is also the pre-step for further studying on thin sections such as mineral identification and measuring the size of minerals. In this paper, a new method for mineral segmentation based on image processing and clustering algorithms is proposed for mineral segmentation in thin sections. In order to segment minerals, using a polarizer microscope, two images in plane and cross polarized lights are captured from each thin sections, and by extracting the color features from the images, minerals inside each thin section are segmented. Therefore, initially, the color features including RGB and HSI components are extracted for each pixels for both images, and then using image processing and clustering algorithms the pixels are clustered and each cluster is related to a segmented mineral. Experimental results indicate that the proposed algorithm produces accurate and reliable results, especially for those thin sections containing altered minerals. The proposed algorithm can be used in such applications as petroleum geology, mineralogy training and NASA mars exploration.

In this paper, an adaptive image contrast enhancement algorithm based on an optimization problem in two dimensional histogram domain is presented. To reduce the unwanted effects of the histogram adjustment, through this optimization-similar to the other methods- the 2D histogram of enhanced image is found in close proximity to input image histogram and uniform distribution, simultaneously. In addition, different from the other methods, by adaptive adjusting the components of a weight matrix, local information is counted. Experimental results in the quantitative and qualitative assessments on a wide range of images demonstrate the performance of the proposed method. Tests have shown that with the addition of the adaptive adjusting the weights, the average performance in contrast enhancement increases 75 and 3 percent from the viewpoint of the AMBE_N and DE_N, respectively.

In this paper we propose a fingerprint image compression method based on the wavelet transform and SPIHT coder. The proposed method employs the proposed technique of dynamic range reduction which benefits from the bimodality of fingerprint images in order to improve the compression efficiency. In addition, we utilized some image enhancement techniques to alleviate the leakage effect as well as further improve the compression efficiency. We have investigated the impacts of compression on recognition efficiency of compressed images. In this investigation we proposed two new measures of breakdown point and downfall slope based on the recognition accuracy curve versus compression bit rate, in order to evaluate the fingerprint image compression approaches more sophisticatedly.Experimental results show that the proposed technique of dynamic range reduction decreased the breakdown point by 0.05 bpp in average. The proposed image enhancement techniques improved the recognition accuracy up to 5% at all compression bit rates higher than the breakdown point. It also decreased the downfall slope. Regarding the PSNR performance, the proposed method outperformed the JPEG2000 and WSQ approaches by 0.8 dB, in average.

Image compression techniquescan bedividedin to two categoriesoflossyandlossless. Predictiveencoder is the basis of many losslessimagecompression methods. Thisencoder predicts the valuesofimage pixelsusing theirneighboringpixelsvalues. The difference between the actual valueand the predictedvalue of each pixelis considered the errorandthese error valuesarecoded. In this paper, a pre-processing methodis proposed tochange the image content arrangement so that the correlations between the neighboring pixelsincrease. By increasing the correlation between neighboring pixels, predictive encoder can more accurately predict the value of each pixel, as a result, entropy is reduced in the error image. According toinformation theory, the lower the imageentropy leads to the higher the capability of theentropy encoderinitscompression. In the proposed method, using the genetic algorithm, an appropriate geometrictransformationof rotationandreflectionis applied on each block of the image to strengthen the correlation between neighboring pixels. In this paper, two compression methods, lossless JPEG and CALIC that are based on predictivecodingareevaluated. The evaluation results of the proposed method on multiple images show that the proposedpre-processingmethodimproves the compressionrate of these two methods.

Image thresholding is a popular method for image segmentation. Histogram is used for image segmentation in image thresholding. In this paper, a multilevel image thresholding is proposed based on teaching-learning-based optimization (TLBO). TLBO is a new population-based metaheuristic inspired by learners and teacher in a classroom. The optimal thresholds are found by maximizing Kapur’s (entropy criterion) thresholding function. The performance of TLBO is explained by considering five images. In addition, the performance is compared with three well known population-based metaheuristics: particle swarm optimization (PSO), genetic algorithm (GA), and differential evolution (DE).Results show that TLBO presents the better performance in terms of fitness value, peak signal to noise ratio (PSNR), Structural-Similarity index (SSIM), and stability.

Face detection is an important part of many computer vision systems and has several applications in areas, such as face tracking, visual surveillance, video conferencing, face recognition, intelligent human-computer interfaces and content-based information retrieval. For use of face detection in this applications, need a fast and precise face detection algorithm. But Detection speed of traditional face detection method based on Ada Boost algorithm is slow since an exhaustive search in image. Over the past few years, the availability of color images with corresponding depth data has increased due to the popularity of low-cost RGB-Depth cameras, notably Kinect. The complementary nature of the depth and visual information provided by the Kinect sensor opens up new opportunities to solve fundamental problems in face detection with intelligently constraining search over the image. In this paper, utilize additional depth data to reduce the computational cost of face detection. Leveraging the additional depth images from a Kinect camera, and use of Recurring in nature idea, we are able to accelerate the Viola-Jones face detector by 270%.