Classificazione di oggetti in immagini attraverso il modello Bag of Visual Words

Generic object recognition is an important function of the human visual system and everybody finds it highly useful in their everyday life. For an artificial vision system it is a really hard, complex and challenging task because instances of the same object category can generate very different images, depending of different variables such as illumination conditions, the pose of an object, the viewpoint of the camera, partial occlusions, and unrelated background clutter. The purpose of this thesis is to develop a system that is able to classify objects in 2D images based on the context, and identify to which category the object belongs to. Given an image, the system can classify it and decide the correct categorie of the object. Furthermore the objective of this thesis is also to test the performance and the precision of different supervised Machine Learning algorithms in this specific task of object image categorization. Through different experiments the implemented application reveals good categorization performances despite the difficulty of the problem. However this project is open to future improvement; it is possible to implement new algorithms that has not been invented yet or using other techniques to extract features to make the system more reliable. This application can be installed inside an embedded system and after trained (performed outside the system), so it can become able to classify objects in a real-time. The information given from a 3D stereocamera, developed inside the department of Computer Engineering of the University of Bologna, can be used to improve the accuracy of the classification task. The idea is to segment a single object in a scene using the depth given from a stereocamera and in this way make the classification more accurate.

The importance of visual and olfactory stimuli during flower visits in Apis mellifera

Flowers attract honeybees using colour and scent signals. Bimodality (having both scent and colour) in flowers leads to increased visitation rates, but how the signals influence each other in a foraging situation is still quite controversial. We studied four basic questions: When faced with conflicting scent and colour information, will bees choose by scent and ignore the “wrong” colour, or vice versa? To get to the bottom of this question, we trained bees on scent-colour combination AX (rewarded) versus BY (unrewarded) and tested them on AY (previously rewarded colour and unrewarded scent) versus BX (previously rewarded scent and unrewarded colour). It turned out that the result depends on stimulus quality: if the colours are very similar (unsaturated blue and blue-green), bees choose by scent. If they are very different (saturated blue and yellow), bees choose by colour. We used the same scents, lavender and rosemary, in both cases. Our second question was: Are individual bees hardwired to use colour and ignore scent (or vice versa), or can this behaviour be modified, depending on which cue is more readily available in the current foraging context? To study this question, we picked colour-preferring bees and gave them extra training on scent-only stimuli. Afterwards, we tested if their preference had changed, and if they still remembered the scent stimulus they had originally used as their main cue. We came to the conclusion that a colour preference can be reversed through scent-only training. We also gave scent-preferring bees extra training on colour-only stimuli, and tested for a change in their preference. The number of animals tested was too small for statistical tests (n = 4), but a common tendency suggested that colour-only training leads to a preference for colour. A preference to forage by a certain sensory modality therefore appears to be not fixed but flexible, and adapted to the bee’s surroundings. Our third question was: Do bees learn bimodal stimuli as the sum of their parts (elemental learning), or as a new stimulus which is different from the sum of the components’ parts (configural learning)? We trained bees on bimodal stimuli, then tested them on the colour components only, and the scent components only. We performed this experiment with a similar colour set (unsaturated blue and blue-green, as above), and a very different colour set (saturated blue and yellow), but used lavender and rosemary for scent stimuli in both cases. Our experiment yielded unexpected results: with the different colours, the results were best explained by elemental learning, but with the similar colour set, bees exhibited configural learning. Still, their memory of the bimodal compound was excellent. Finally, we looked at reverse-learning. We reverse-trained bees with bimodal stimuli to find out whether bimodality leads to better reverse-learning compared to monomodal stimuli. We trained bees on AX (rewarded) versus BY (unrewarded), then on AX (unrewarded) versus BY (rewarded), and finally on AX (rewarded) and BY (unrewarded) again. We performed this experiment with both colour sets, always using the same two scents (lavender and rosemary). It turned out that bimodality does not help bees “see the pattern” and anticipate the switch. Generally, bees trained on the different colour set performed better than bees trained on the similar colour set, indicating that stimulus salience influences reverse-learning.

Rapid adaptation of visual search in simulated hemianopia

Patients with homonymous hemianopia have altered visual search patterns, but it is unclear how rapidly this develops and whether it reflects a strategic adaptation to altered perception or plastic changes to tissue damage. To study the temporal dynamics of adaptation alone, we used a gaze-contingent display to simulate left or right hemianopia in 10 healthy individuals as they performed 25 visual search trials. Visual search was slower and less accurate in hemianopic than in full-field viewing. With full-field viewing, there were improvements in search speed, fixation density, and number of fixations over the first 9 trials, then stable performance. With hemianopic viewing, there was a rapid shift of fixation into the blind field over the first 5-7 trials, followed by continuing gradual improvements in completion time, number of fixations, and fixation density over all 25 trials. We conclude that in the first minutes after onset of hemianopia, there is a biphasic pattern of adaptation to altered perception: an early rapid qualitative change that shifts visual search into the blind side, followed by more gradual gains in the efficiency of using this new strategy, a pattern that has parallels in other studies of motor learning.

Three-year results of visual outcome with disease activity-guided ranibizumab algorithm for the treatment of exudative age-related macular degeneration

PURPOSE To evaluate 3-year follow-up treatment outcomes with ranibizumab (Lucentis(®)) 0.5 mg administered either monthly or quarterly on a pro re nata (PRN) basis according to a disease activity-guided monitoring and treatment algorithm. METHODS A total of 316 treatment-naive eyes of 316 patients with exudative age-related macular degeneration met the criteria for inclusion in this retrospective, interventional case series. Patients were treated with ranibizumab 0.5 mg according to a disease activity-guided algorithm with monthly monitoring. Optical coherence tomography and fluorescein angiography were routinely used to assess disease activity: active lesions were treated with a series of three monthly injections, whereas inactive lesions were treated with quarterly injections. RESULTS Mean Early Treatment Diabetic Retinopathy Study best-corrected visual acuity improved from 52 letters at baseline to 59 letters at 12 months, achieved with a mean of 7.1 injections, 61 letters at 24 months with a mean of 5.0 injections administered in the second year and 60 letters at 36 months with a mean number of 5.2 injections. CONCLUSIONS Monthly visits and a morphology-driven PRN regimen with 3 injections in case of recurrence plus quarterly injections in case of inactive CNV resulted in an average VA gain of 7-9 letters that could be maintained over 3 years.

Misleading Population Estimates: Biases and Consistency of Visual Surveys and Matrix Modelling in the Endangered Bearded Vulture

Conservation strategies for long-lived vertebrates require accurate estimates of parameters relative to the populations' size, numbers of non-breeding individuals (the “cryptic” fraction of the population) and the age structure. Frequently, visual survey techniques are used to make these estimates but the accuracy of these approaches is questionable, mainly because of the existence of numerous potential biases. Here we compare data on population trends and age structure in a bearded vulture (Gypaetus barbatus) population from visual surveys performed at supplementary feeding stations with data derived from population matrix-modelling approximations. Our results suggest that visual surveys overestimate the number of immature (<2 years old) birds, whereas subadults (3–5 y.o.) and adults (>6 y.o.) were underestimated in comparison with the predictions of a population model using a stable-age distribution. In addition, we found that visual surveys did not provide conclusive information on true variations in the size of the focal population. Our results suggest that although long-term studies (i.e. population matrix modelling based on capture-recapture procedures) are a more time-consuming method, they provide more reliable and robust estimates of population parameters needed in designing and applying conservation strategies. The findings shown here are likely transferable to the management and conservation of other long-lived vertebrate populations that share similar life-history traits and ecological requirements.

Longitudinal testing of visual perception in dementia with Lewy bodies and Alzheimer's disease

Visuo-perceptual abnormalities are a prominent feature in dementia with Lewy bodies (DLB) and also occur in Alzheimer's disease (AD) to a lesser extent. We studied the progression of visuo-perceptual abnormalities over a 12-month period in DLB and AD by using a novel computerised test battery.

[Simultaneous defect of visual fields and loss of libido--a coincidence?]

We present the case of a 60 year old male patient with incidentally detected visual abnormalities. Detailed personal history revealed a hypogonadism that had been present for several years. Further investigations established the diagnosis of an infiltrative macroadenoma. Medical treatment with cabergoline led to a rapid regression of ophthalmologic symptoms and, subsequently, of tumor size. In male subjects symptoms of hypogonadism are often reported only late in the course of the disease, thereby leading to a generally larger tumor size at the point of diagnosis. In contrast to other pituitary tumors that are mainly treated by surgery, medical treatment with dopamine agonists is the principal therapeutic option in prolactinomas.

Functional imbalance of visual pathways indicates alternative face processing strategies in autism

OBJECTIVE: To investigate whether autistic subjects show a different pattern of neural activity than healthy individuals during processing of faces and complex patterns. METHODS: Blood oxygen level-dependent (BOLD) signal changes accompanying visual processing of faces and complex patterns were analyzed in an autistic group (n = 7; 25.3 [6.9] years) and a control group (n = 7; 27.7 [7.8] years). RESULTS: Compared with unaffected subjects, autistic subjects demonstrated lower BOLD signals in the fusiform gyrus, most prominently during face processing, and higher signals in the more object-related medial occipital gyrus. Further signal increases in autistic subjects vs controls were found in regions highly important for visual search: the superior parietal lobule and the medial frontal gyrus, where the frontal eye fields are located. CONCLUSIONS: The cortical activation pattern during face processing indicates deficits in the face-specific regions, with higher activations in regions involved in visual search. These findings reflect different strategies for visual processing, supporting models that propose a predisposition to local rather than global modes of information processing in autism.

Recovery of visual field and acuity after removal of epiretinal and inner limiting membranes

BACKGROUND: Visual acuity serves as only a rough gauge of macular function. The aim therefore was to ascertain whether central an assessment of the central visual field afforded a closer insight into visual function after removal of epiretinal membranes and Infracyanine-Green- or Trypan-Blue-assisted peeling of the inner limiting membrane. Patients and methods: Fourty-three patients undergoing pars-plana vitrectomy for the removal of epimacular membranes and dye-assisted peeling of the inner limiting membrane using either Infracyanine Green (n = 29; group 1) or Trypan Blue (n = 14; group 2) were monitored prospectively for 12 months. Preoperatively, and 1, 6 and 12 months postoperatively, distance and reading visual acuities were evaluated; the central visual field was assessed by automated static perimetry. RESULTS: Twelve months after surgery, distance and reading visual acuities had improved in both groups, but to a significant degree only in Trypan-Blue-treated eyes. The difference between the two groups was not significant. Likewise at this juncture, the mean size of the visual-field defect remained unchanged in Trypan-Blue-treated eyes (preoperative: 4.3 (SD 2.1) dB; 12 months: 4.0 (2.1) dB (p = 0.15)), but had increased in Infracyanine-Green-treated ones (from 5.3 (3.7) dB to 8.0 (5.2) dB (p = 0.027)). CONCLUSION: Unlike visual acuity, the central visual field had deteriorated in Infracyanine-Green-treated eyes but not in Trypan-Blue-treated eyes 12 months after surgery. Hence, as a predictor of functional outcome, testing of the central visual field may be a more sensitive gauge than visual acuity. Furthermore, Infracyanine Green may have a chronic and potentially clinically relevant effect on the macula which is not reflected in the visual acuity.

Comparison of visual-tactile, radiographic, and histologic diagnoses of subgingival crown margin caries- an in vitro study

The purpose of this study was to investigate the accuracy of diagnosing interproximal subgingival caries at crown margins. A total of 32 subgingival interproximal crown margin areas were examined by 10 clinicians (n = 320) using conventional diagnostic methods on extracted, crowned teeth mounted in a specially designed cast. Crown margins were located 1.5 mm below the level of the artificial gingiva. Clinical and radiographic diagnoses were compared to the histopathologic findings for each site. Both visual-tactile and radiographic evaluations revealed a weak diagnostic accuracy for interproximal subgingival crown margin caries.